The genealogical tree of ethanol: gas-phase formation of glycolaldehyde, acetic acid and formic acid

Despite the harsh conditions of the interstellar medium, chemistry thrives in it, especially in star forming regions where several interstellar complex organic molecules (iCOMs) have been detected. Yet, how these species are synthesised is a mystery. The majority of current models claim that this happens on interstellar grain surfaces. Nevertheless, evidence is mounting that neutral gas-phase chemistry plays an important role. In this article, we propose a new scheme for the gas-phase synthesis of glycolaldehyde, a species with a prebiotic potential and for which no gas-phase formation route was previously known. In the proposed scheme, the ancestor is ethanol and the glycolaldehyde sister species are acetic acid (another iCOM with unknown gas-phase formation routes) and formic acid. For the reactions of the new scheme with no available data, we have performed electronic structure and kinetics calculations deriving rate coefficients and branching ratios. Furthermore, after a careful review of the chemistry literature, we revised the available chemical networks, adding and correcting several reactions related to glycolaldehyde, acetic acid and formic acid. The new chemical network has been used in an astrochemical model to predict the abundance of glycolaldehyde, acetic acid and formic acid. The predicted abundance of glycolaldehyde depends on the ethanol abundance in the gas phase and is in excellent agreement with the measured one in hot corinos and shock sites. Our new model overpredicts the abundance of acetic acid and formic acid by about a factor of ten, which might imply a yet incomplete reaction network.


INTRODUCTION
About one third of the ca. 200 molecules detected in the Inter-Stellar Medium (ISM) are constituted by six or more atoms (source http://www.astro.unikoeln.de/cdms/molecules). All these molecules contain at least one carbon atom. In the following, we will call iCOMs 1 (for interstellar Complex Organic Molecules) molecules with at least six atoms and containing at least one carbon atom. In the majority of iCOMs, hydrogen and oxygen are the additional elements. To date, slightly more than forty iCOMs which contain elements other than C and H have been detected. Thus, they represent about 20% of detected ISM molecules.
Even though the presence of iCOMs has been known for decades (for instance formamide has been detected in 1971 by Rubin et al.), the processes that lead to their synthesis are still hotly debated. Specifically, it is nowadays often assumed that iCOMs are mostly synthesised on grain surfaces during the so-called warm-up phase, when various radicals trapped in the grain mantles acquire mobility and recombine into large molecules (e.g. Garrod & Herbst 2006;Garrod et al. 2008). Yet, recent detections of iCOMs in cold environments (Bacmann et al. 2012;Cernicharo et a. 2012;Jaber et al. 2014;Vastel et al. 2014;Jiménez-Serra et al. 2016) have challenged this exclusive role of grain-surface chemistry. Clearly, some gas-phase chemistry is at work in cold environments (Vasyunin & Herbst 2013;Balucani et al. 2015;Ruaud et al. 2016;Vasyunin et al. 2017).
Supporting the idea that grain-surface chemistry cannot synthesise all detected iCOMs, recent quantum chemistry calculations have shown that the combination of radicals trapped in amorphous water ice does not necessarily lead to larger molecules (Enrique-Romero et al. 2016;Rimola et al. 2017), in particular to iCOMs, as predicted by the above mentioned grain-surface chemical models. The basic reason is that radicals are not oriented in a way for the recombination reaction to occur, as they are trapped by the water-ice molecules in a configuration that favors other two-product reactions.
Following up with the idea that gas-phase reactions might have been overlooked, Kahane et al. (2013) proposed that formamide (NH 2 CHO) is formed by the reaction of formaldehyde (H 2 CO) and amidogen (NH 2 ). Barone et al. (2015), Vazart et al. (2016) and Skouteris et al. (2017) carried out theoretical computations showing that this reaction can efficiently occur at 1 Please note that we added i to the commonly used COMs acronym in order to be clear that these molecules are only complex in the interstellar context , contrary to what chemists would consider complex in the terrestrial context. low temperatures (contrarily to what claimed by Song & Kastner, 2016) and can explain the available observations. Other studies have explored possible gas-phase ion-neutral reactions leading to formamide (Spezia et al. 2016). More recently, an observational study obtained with the IRAM-NOEMA interferometer provided additional support to the gas-phase formation of formamide . Similarly, observations of its deuterated forms by Coutens et al. (2016) are in good agreement with the theoretical predictions of deuterated amidogen or formaldehyde leading to deuterated formamide .
In summary, it is well possible that gas-phase reactions play an important role in the formation of iCOMs, but more systems need to be studied to understand their real importance.
The aim of this paper is to understand whether previously overlooked gas phase routes can lead to glycolaldehyde (HCOCH 2 OH), a "special" iCOM because of its prebiotic potential. Glycolaldehyde has been detected towards the giant molecular cloud complex SgrB2 (Hollis et al. 2000) towards high-and low-mass star forming regions (Beltran et al. 2009;Jorgensen et al. 2012Jorgensen et al. , 2016Coutens et al. 2015;Taquet et al. 2015;De Simone et al. 2017) and in shocked regions (Lefloch et al. 2017). Several mechanisms of glycolaldehyde synthesis on grain surfaces were proposed: they involve recombination of radicals (Garrod et al. 2008), UV or particle irradiation (Woods et al. 2012;Maity et al. 2014;Fedoseev et al. 2015;Butscher et al. 2015, Chuang et al. 2017). On the contrary, previous work on glycolaldehyde formation in the gas phase could not identify plausible interstellar routes (Wang & Bowie 2010;Jalbout 2007).
In a recent work, Lefloch et al. (2017) showed that there is a correlation between the abundances of glycolaldehyde and ethanol, even though this is based on only four sources. Following the suggestion provided by this possible correlation, in this work we propose a series of gas-phase reactions that start from ethanol (CH 3 CH 2 OH) and that lead to the synthesis of glycolaldehyde in a sequence of gas-phase reactions which is similar to that connecting dimethyl ether and methyl formate, two other common iCOMs . As in that case, the sequence of reactions starts with the conversion of ethanol (or its isomer dimethyl ether in the case analyzed by Balucani et al. 2015) to a reactive radical that can further react with abundant O atoms leading to glycolaldehyde (methyl formate in the case analyzed by Balucani et al. 2015). Some of the necessary data to test this hypothesis were available in the literature (see Sec. 3). Some crucial data were instead missing. Therefore, to verify whether the proposed route is efficient in the ISM conditions, we have performed dedicated electronic structure and kinetics calculations for the missing reactions. The main result of this work is that ethanol can be considered not only the ancestor of glycolaldehyde, but also of formic acid (HCOOH) and acetic acid (CH 3 COOH), another common iCOM. In addition, ethanol is revealed to be one of the precursors of acetaldehyde, a widely spread iCOM. In other words, ethanol can be considered the progenitor of three iCOMs and of formic acid.
The article is organised as follows. In Section 2, we briefly review the previously known gas-phase reactions of the above iCOMs and formic acid. We then present the overall scheme and justification of the newly proposed reactions in Section 3 and the employed methodology and results of our computations in Section 4. In Section 5 we present the predictions obtained by an astrochemical model including the new reactions and we discuss the comparison with observations. Section 6 concludes this article.

Glycolaldehyde (HCOCH 2 OH)
No reactions are reported in the KIDA or UMIST databases. Halfen et al. (2006) proposed that protonated formaldehyde (H 2 COH + ) could react with formaldehyde to produce protonated glycolaldehyde in a radiative association reaction. The electron recombination of protonated glycolaldehyde then ends in glycolaldehyde by losing an H atom. However, electronic structure calculations by Horn et al. (2004) showed that the relative association products do not have the molecular structure of protonated glycolaldehyde. Furthermore, it has been known since the experimental work by Karpas & Klein (1975) that two-product exothermic channels are available for the H 2 COH + + H 2 COH reaction, which strongly reduces the probability of radiative association in the absence of secondary collisions. Finally, Woods et al. (2012Woods et al. ( , 2013 claimed that this route is inefficient and cannot reproduce the observed abundances.

Acetic acid (CH 3 COOH)
No reactions are reported in the KIDA or UMIST databases for this species and we are not aware of proposed schemes of its formation in the gas phase.

Formic acid (HCOOH)
KIDA lists one reaction, CH 3 O + 2 + e − , which is assumed to produce 50% of HCOOH (+ H) and 50% of CO 2 (+ H 2 + H). In UMIST, the same reaction, reported as HCOOH + 2 + e − , is globally faster, but has a branching ratio of only 13% for the HCOOH channel, with the major channel leading to HCO + OH + H. The UMIST rate coefficients are based on the experiments by Vigren et al. (2013), who were, however, only able to demonstrate that heavy products with at least one C and two O atoms account for 13% of the global reaction. This could include also CO 2 formation, as suggested in KIDA. In addition to that, there are issues concerning the formation of the HCOOH + 2 / CH 3 O + 2 isomers. The main formation route of so-called protonated formic acid is considered to be the radiative association reaction HCO + + H 2 O (Herbst 1985). In the KIDA and UMIST networks, this process is present with a relatively high rate coefficient of 1.7 × 10 −12 cm 3 s −1 at 100 K. In general, the rate coefficients of most radiative association reactions are poorly defined and can only be estimated. From what is known so far, a significant probability for radiative association reactions to occur can be expected only when there are no exothermic two-product channels or when the presence of very high exit barriers prevents a fast escape from the potential well associated to the addition intermediate. Only in these cases, indeed, the lifetime of the intermediate can be long enough to permit the spontaneous emission of photons necessary to its stabilization (in the absence of ternary collisions, as in interstellar environments). As warned by Herbst (1985), this is not the case of the HCO + + H 2 O which is indeed a fast reaction (3.64 × 10 −9 cm 3 s −1 at 100 K) with a very exothermic two-product channel (leading to CO + H 3 O + ). For this reason, we have deleted the radiative association of HCO + + H 2 O from our reaction network. The other routes of HCOOH + 2 / CH 3 O + 2 formation have already been proved to be marginal (Vigren et al. 2013). In addition, UMIST also reports OH + H 2 CO → HCOOH + H, a reaction which has been widely studied at higher temperatures. Since the channel leading to HCOOH + H has an entrance barrier of 23.8 kJ/mol (Xu & Lin 2007), we did not consider it in our network. Other reactions listed in UMIST are expected to make a negligible contribution to HCOOH formation.

Acetaldehyde (CH 3 CHO)
Several reactions forming acetaldehyde are reported in KIDA. A possibly major formation route is the electron recombination of protonated acetaldehyde which ends up in acetaldehyde. However, the protonated acetaldehyde is mostly formed by the reaction CH 3 OCH 3 + H + → CH 3 CHOH + + H 2 , which would imply a substantial and improbable rearrangement of the nuclei. We dropped, therefore, this reaction from the network. Another major formation reaction route, and often the most important one in several published models, involves atomic oxygen and ethyl radical: CH 3 CH 2 + O → CH 3 CHO + H (Charnley et al. 2004;Vastel et al. 2014).
3. NEW REACTION SCHEME Figure 1 illustrates the scheme of the reactions proposed in this work.
In the proposed new reaction scheme, ethanol is chemically activated by one of the abundant atomic or molecular radicals that are present in interstellar clouds. In Balucani et al. (2015), several radicals were considered and atomic chlorine was suggested as the major contributor in converting dimethyl ether into the reactive methylmethoxy radical. Vasyunin et al. (2017) considered, instead, that the abundant OH radicals are the major players by referring to the recent work by Shannon et al. (2014), who performed kinetics experiments at T as low as 60 K 2 .
We follow the same approach here, that is, we have considered the reaction of ethanol with Cl atoms and OH radicals as the initiating steps. In the following we give details of the employed reactions.
3.1. The initiating reaction Cl + CH 3 CH 2 OH This reaction has been widely investigated at room or higher temperatures. In particular, Taatjes et al. (1999) were able to derive the H-abstraction site-specific rate coefficients as, differently from the case of dimethyl ether, there are three different kinds of hydrogen atoms that Cl (or other radicals) can abstract: (i) three equivalent H atoms from the methyl group (CH 3 ), (ii) two equivalent H atoms from the methylene group (CH 2 ) and (iii) one H atom from the hydroxyl group (OH). According to the measurements by Taatjes et al. (1999) at room temperature, H abstraction from the methylene group (ii) is by far the dominant pathway accounting for about 90% of the total reaction. H abstraction from the methyl group (i) accounts for the rest, while abstraction of the hydroxyl hydrogen (iii) is negligible. Unfortunately, there are no experimental data at the low temperatures of interest in our case. Therefore, in our network we have included the reaction channels (labelled reaction 1 and 2 in Table 1) with their room temperature values. In addition, the product branching ratio for the reaction Cl + CH 3 CH 2 OH could slightly vary with the temperature as channel (1) is exothermic by 44.8 kJ/mol, channel (2) by 17.6 kJ/mol, while the channel leading to CH 3 CH 2 O is slightly endothermic by 3.8 kJ/mol (Rusic et al. 1994). Moreover, electronic structure calculations by Rudic et al. (2002) predicted a very small barrier of ca. 3 kJ/mol for channel (2).

The initiating reaction OH + CH 3 CH 2 OH
This reaction has been widely investigated at room or higher temperatures and, more recently, also at temperatures as low as 50 K by Caravan et al. (2015). Similarly to the case of the analogous reaction with dimethyl ether in low temperature experiments, the observed pressure dependence of the rate coefficients provided evidence that, in addition to the bimolecular abstraction channel leading to products, collisional stabilization of the weakly bound OH-ethanol complex occurred under their experimental conditions − thus providing an artificially high rate coefficient (we remind that such a mechanism cannot be present under the rarefied conditions of the interstellar medium). In our network, we have considered the value of the rate coefficient that Caravan et al. (2015) recommended as representative of the sole twoproduct channel, that is, at 82-91 K, 2.7 (± 0.8)×10 −11 cm 3 s −1 . Also in this case the H-abstraction can occur at three different sites, leading to the radicals CH 2 CH 2 OH, CH 3 CHOH and CH 3 CH 2 O. As CH 3 O (+ H 2 O) was determined to be the major product in the analogous reaction OH + CH 3 OH by Shannon et al. (2013), Caravan et al. (2015) attempted the detection of CH 3 CH 2 O, but failed. At higher T, the formation of CH 3 CHOH (+ H 2 O) is known to be the main channel, with a branching ratio varying between 0.75−0.9, while the CH 2 CH 2 OH (+ H 2 O) channel accounts for the rest (Marinov, 1999;Carr et al. 2011). The branching ratio was also seen to vary with T. We have, therefore, decided to test two different scenarios: in the first scenario, we have assumed a value of 0.9:0.1 (reactions 3a and 4a in Table 1); in the second scenario, we have assumed a value of 0.7:0.3 (reactions 3b and 4b in Table 1). Even though these seem to be reasonable ranges, a final value could be adopted only when low T determination of the branching ratio becomes available. 3.3. Second step: reactions of CH 3 CHOH and CH 2 CH 2 OH with O Further reactions of the radicals produced in the initiating steps with atomic oxygen generate the species shown in Fig. 1. To the best of our knowledge, only fragmentary data were available in the literature concerning the reaction channel O + CH 3 CHOH → CH 3 CHO + OH (Edelbuttel-Einhaus et al., 1992) , so we have performed dedicated electronic structure and kinetics calculations. The results of the calculations are reported in Section 4.
The reactions with their branching ratios and rate coefficients are reported in Table 1.

Electronic structure calculations
Calculations have been performed with a development version of the Gaussian suite of programs (Frisch et al. 2013) as well as with the CFOUR program package. 3 Geometry optimizations for all stationary points were performed with the double-hybrid B2PLYP functional (Grimme 2006) in conjunction with the m-aug-cc-pVTZ basis set (Papajak et al., 2009;Dunning 1989) where d functions on hydrogens have been removed. Semiempirical dispersion contributions were also included by means of the D3BJ model of Grimme Grimme et al. 2011). Full geometry optimizations have been performed for all molecules checking the nature of the obtained structures (minima or first order saddle points) by diagonalizing their Hessians. For each stationary points, the anharmonic force field has been computed at the B2PLYP-D3BJ/m-aug-cc-pVTZ level in order to evaluate the zero-point energies (ZPEs) using vibrational perturbation theory (VPT2). To obtain accurate electronic energies, the coupled-cluster singles and doubles approximation augmented by a perturbative treatment of triple excitations (CCSD(T), Raghavachari et al. 1989) was employed in conjunction with extrapolation to the complete basis set limit (CBS) and inclusion of core-correlation effects (CV), thus leading to the so-called CCSD(T)/CBS+CV approach (Heckert et al. 2005, Heckert et al. 2006. The cc-pVnZ, with n=T,Q, basis sets (Dunning 1989) were used in the extrapolation to the CBS limit, while the cc-pCVTZ set (Woon & Dunning 1995) was employed for evaluating the CV correction.

Reaction paths for the reactions O + CH 2 CH 2 OH and O + CH 3 CHOH
The reaction paths for both schemes are shown in Figures 2, 3. Figure 2 exhibits the reaction path following the O( 3 P) + CH 2 CH 2 OH addition. The barrierless addition of oxygen leads to the (I) intermediate, which is about 400 kJ/mol more stable than the reactants. Its trans counterpart, the slightly less stable (by 8 kJ/mol) intermediate (II), can easily be reached from the cis species through a 20 kJ/mol barrier (TS1). Both intermediates are then able to undergo dissociation to formaldehyde and CH 2 OH through the transition states (TS6) and (TS5). These dissociations exhibit barriers around 55 kJ/mol. Other dissociations can also be observed from both these intermediates, leading to cis-or trans-glycolaldehyde and H, respectively through (TS2) and (TS3) that are about 115 kJ/mol higher in energy than their corresponding dissociating intermediates. The most stable products that can be obtained with this path are formaldehyde + CH 2 OH, with a relative energy of around −346 kJ/mol followed by the glycolaldehyde isomers, with a relative energy of around −310 kJ/mol. Wang and Bowie (2010) performed electronic structure calculations for the same global potential energy surface, but considering the possible formation of glycolaldehyde from the reaction H 2 CO + CH 2 OH. Their entrance channel, therefore, is one of the present exit channels in our case. The energy values of the corresponding  Taatjes et al. (1999). 2 We have adopted the total value measured by Caravan et al (2015) in the range 82−91 K, partitioned according to two possible scenarios for the channels 3 and 4 (see text). 3 The rate coefficients and product branching ratios are those computed in the present work.  intermediates and transition states can be compared and are in general good agreement. More specifically, our values for the energy of the transition states and global endothermicity are lower than the ones presented in that work. Nevertheless, both are still high enough to render the reaction H 2 CO + CH 2 OH → HCOCH 2 OH + H prohibitive in the ISM. Figure 3 exhibits the reaction path following the O( 3 P) + CH 3 CHOH addition. The barrierless addition of oxygen leads to the (III) intermediate, which is about 410 kJ/mol more stable than the reactants. This species is then able to undergo a dissociation into formic acid and the CH 3 radical, through the (TS11) transition state. This dissociation exhibits an approximate 45 kJ/mol barrier. Another dissociation can also be observed, leading to acetic acid and H, through (TS9) which is about 60 kJ/mol higher in energy than (III). Starting again from (III), elimination of an OH radical can occur (through the (TS8) transition state, lying approximately 100 kJ/mol above the intermediate) yielding acetaldehyde. Furthermore, from (III) hydrogen migration can lead to compound (IV) with a barrier of 105 (TS7) kJ/mol. The intermediate (IV) is found to be around 460 kJ/mol more stable than the precursor. (IV) can also undergo hydrogen loss resulting in acetic acid and H through a 117 kJ/mol barrier (TS10). The most stable products are formic acid and CH 3 , with a relative energy of −419.5 kJ/mol with respect to the precursors, followed by acetic acid + H that exhibits a relative energy of −395.3 kJ/mol.

Kinetics calculations
As in previous work (Balucani et al. 2012, Leonori et al. 2013, Sleiman et al. 2017) a combination of capture theory and the Rice-Ramsperger-Kassel-Marcus (RRKM) calculations was used to determine the relevant rate coefficients and branching ratios. For the first steps (addition of the O( 3 P) atom to either the CH 2 CH 2 OH or the CH 3 CHOH radicals) capture theory was used, whereas for the subsequent reactions energy-dependent rate constants were calculated using the RRKM scheme and taking into account anharmonicity of the vibrational modes. Subsequently, the master equation was solved at all relevant energies for both systems (to take into account the overall reaction scheme), Boltzmann averaging was carried out to obtain temperature-dependent rate coefficients and, finallly, rate coefficients were fitted to the form k = α × (T gas /300K) β × exp[−γ/T gas . The values of α, β and γ in each case are given in Table 1.
Back-dissociation is negligible in both cases, due to the high stability of the initial intermediate and the pres-ence of very exothermic channels with low exit energy barriers. When the O atom adds to CH 2 CH 2 OH, the most probable fates of the radical are cleavage of the C-C bond (to yield formaldehyde and the CH 2 OH radical) or an H atom elimination to give glycolaldehyde. The first one dominates at all temperatures due to the significantly lower energy barrier involved. Nevertheless, as can be seen from the final values, a substantial percentage of the intermediate goes to glycolaldehyde (yield of 19%).
An analogous situation presents itself when the O atom adds to the CH 3 CHOH radical. The two most probable fates of the radical are C-C bond cleavage, yielding HCOOH and a methyl radical and elimination of an H atom from the α C atom to yield acetic acid (CH 3 COOH). Before the elimination, there is also the possibility of an H atom transfer from the α C atom to the newly added oxygen, followed by an H atom elimination from one of the two O atoms. However, this is the least followed path towards acetic acid formation, both because of the higher barrier involved and the longer reaction path. The most abundant product is formic acid (with a yield of 59%), because of the lower barrier involved, while acetic acid formation is second highest (33.5%). Finally, there is some possibility of elimination of an OH radical from the initial intermediate, yielding CH 3 CHO. The barrier for this process is considerably higher than both previous ones and therefore the rate of formation of acetaldehyde is lower. Concerning this last point, it should be noted that Edelbüttel-Einhaus et al. (1992) were, instead, only able to detect the CH 3 CHO product in their room temperature experiments.

Description of the model
In order to understand whether the proposed new reaction scheme and rate coefficients can explain observations towards hot corinos, we used an astrochemical model which simulates their conditions. Towards this scope, we used a modified version (to improve its versatility) of the time-dependent chemistry code NA-HOON 4 . We run the code in two steps. In the first step, we follow the chemical composition of the molecular cloud from which the hot corino evolves. We start with the standard atomic state with the element abundances listed in Table 2 and wait for the steady state composition of the gas. We then simulate the hot corino appearance by injecting into the gas phase the species pre-viously frozen into the dust grain mantles in the quantities listed in Table 2. The assumption is that when the dust grain temperature reaches ∼100 K the ice mantles sublimate and all species trapped in the water matrix co-desorb with it. This is a rough approximation, but enough for the scope of this article, the aim of which is to provide an order of magnitude of the species abundances.
The starting point of our proposed chemical scheme is the sublimation of ethanol from interstellar grains. Since IR observations have (only) possibly identified this species in the ice mantles of interstellar grains (Boogert et al. 2015, Schutte et al., 1999Oberg et al., 2011), its abundance is a parameter of the model, that we varied between 10 −8 and 10 −6 .
Similarly, it is not clear what the abundance of atomic chlorine in hot corinos is. On Earth, chlorine is mostly in oceans and very little in rocks. Therefore, only a little fraction of chlorine is probably contained in the refractory grains of the ISM (Jenkins et al. 2009). Observations of HCl in hot cores/corinos and shock sites show that this molecule has an abundance ∼ 10 −9 (e.g. Peng et al. 2010;Codella et al. 2012;Kama et al. 2015), namely about 300 times lower than the solar abundance. Since in hot cores/corinos and shock sites the grain mantles components are injected into the gas phase, these observations show that HCl is not the major reservoir of chlorine, contrarily to model predictions (e.g. Neufeld et al. 2012). It is, therefore, possible, if not likely, that a large fraction of Cl is atomic. Since no observations exist to constrain the abundance of atomic chlorine in hot corinos, its abundance is considered a parameter of the model and it is varied between 10 −9 and 10 −7 . The highest value corresponds to the assumption that 70% of Cl is depleted in the refractory grains or in other Clbearing molecules.
The other crucial species involved in the initiating steps of the proposed scheme is OH (Table 1). This radical is a product of the injection of water from the ice mantles and it is self-consistently computed by the astrochemical model. In this case, we adopted the "standard" value for injected water of 1 × 10 −4 (e.g. Boogert et al. 2015), quoted in Table 2.
Finally, the H density of the hot corinos is assumed to be 2 × 10 8 cm −3 and its temperature 100 K. The cosmic ray ionisation rate is assumed to be 3 × 10 −16 s −1 (e.g. Caselli & Ceccarelli 2012).

Species Abundance
Step 1: elemental abundances Mg 7.0×10 −9 Fe 3.0×10 −9 Na 2.0×10 −9 Cl 1.0×10 −9 F 1.0×10 −9 Step 2: injected mantle species Cl 1.0×10 −9 -1.0×10 −7  Boogert et al. (2015), who give them relative to H2O. The elemental abundances are 5% the solar ones for oxygen, carbon and nitrogen, and 0.5% for the heavier elements, to account for the freeze-out of these elements in the molecular cloud. Please note that the Cl abundance is a parameter of the model (see text). All abundances are with respect to H-atoms.
We used the KIDA network 5 , modified following Loison et al. (2014), Balucani et al. (2015) and Skouteris et al. (2017), plus the reactions in Tables 1 and 3. The first table reports the reactions in the new proposed scheme (Section 3), whereas the second table lists the reactions added to complete the formation and destruction routes of the newly introduced species (i.e. not present in the KIDA database) or the reactions that were modified with respect to the KIDA content. Notes with the relevant references and arguments are listed in the Appendix.
We emphasise that the first step, which leads to CH 3 CHOH and CH 2 CH 2 OH from ethanol, can be obtained either via the reaction with atomic Cl or OH. Since in the literature there have been different values of branching ratios for the latter reaction, we considered two different scenarions (see Sec. 3), namely the two sets of reactions (3a), (4a), and (3b), (4b), respectively.

Results
We run three grids of models, each with the abundance of injected ethanol and atomic chlorine in the range reported in Table 2. The three grids are obtained by varying the conditions of the initiating steps of our proposed scheme (Fig. 1), namely the reactions of Table 1 numbered 1 to 4. In the first grid, we adopted the rates of (3a) and (4a), in the second grid the rates of (3b) and (4b), and in the last grid we did not consider the reactions (3) and (4), to quantify the role of atomic chlorine.
The results at 1.5×10 3 yr (the approximate age of hot corinos and of L1157-B1), are shown in Fig. 4. The figure shows the abundance of glycolaldehyde as a function of the ethanol abundance in the gas, which can be different from the one injected from the mantles (as it is used to make the other species). We predict glycolaldehyde abundances in the range of 3 × 10 −10 to 2 × 10 −8 , and ethanol from 10 −9 to 10 −7 . The largest glycolaldehyde abundances are obtained adopting the most favorable branching ratio of the reaction between ethanol and OH, namely the rates (3b) and (4b) of Table 1. Ignoring the reactions (3) and (4) results in the lowest predicted glycolaldehyde abundances. This means that the ethanol reaction with Cl plays a minor role in our model, provided that a large abundance of water is present. Figure 5 shows the abundance of glycolaldehyde, acetic acid, formic acid and ethanol as a function of time.
In these computations, we adopted an abundance of injected ethanol and atomic chlorine equal to 2.8 × 10 −8 and 2.2 × 10 −8 , respectively. In the conditions assumed by the model (Section 5.1), the injected ethanol is all consumed in about 2000 yr. Formic acid is the one that benefits most, followed by acetic acid and, finally, glycolaldehyde. Before sublimated ethanol is fully consumed, the abundance ratios are HCOOH/CH 3 COOH∼1.5 and CH 3 COOH/HCOCH 2 OH∼10, and are mostly governed by the branching ratios of the first two steps of the proposed reactions ( Fig. 1 and Table 1). Once ethanol is fully consumed, the relative abundance ratios are dominated by the destruction reactions (Table 3). Figure 4. Abundance of glycolaldehyde as a function of ethanol abundance in the gas phase, which can be different from the one injected from the mantles. The computations refer to a gas with temperature equal to 100 K, H nuclei density 2 × 10 8 cm −3 , cosmic ray ionisation rate 3 × 10 −1 s −1 and time 1.5 × 10 3 yr. The three curves refer to models with different reactions of Table 1: adopting the reactions (3b) and (4b) (solid line), reactions (3a) and (4a) (dashed line) and excluding the reaction with the OH radicals (dotteddashed line). The atomic chlorine abundance is 2.2 × 10 −8 in the computations. Measured abundances towards NGC1333, IRAS4A and IRAS2A, IRAS16293-2422 and L1157-B1 are also reported with their uncertainties. Figure 5. Abundance of glycolaldehyde (red), acetic acid (purple), formic acid (cyan) and ethanol (green) as a function of time. The computations are obtained for a gas with temperature equal to 100 K, H nuclei density 2 × 10 8 cm −3 , cosmic ray ionisation rate 3 × 10 −1 s −1 and assuming reactions (3b) and (4b) of Table 1. The abundance of injected ethanol and chlorine is 2.8 × 10 −8 and 2.2 × 10 −8 , respectively. The black dashed line shows the time used to obtain Fig. 4, namely 1500 yr. Table 4 reports the measured abundances of the ethanol, glycolaldehyde, acetic acid and formic acid in hot corinos (NGC1333 IRAS4A and IRAS2A, and IRAS16293-2422Coutens et al. 2015;Jørgensen et al. 2012;Jaber et al. 2014) and  The comparison of the model predictions with measurements of the glycolaldehyde abundance is reported in Fig. 4. When considering that the predictions shown in the figure are obtained for generic hot corino conditions, the agreement with the observations is very encouraging. Obviously, the model with the reactions set (3b), (4b) produces a larger amount of glycolaldehyde (and a lower one of formic acid and acetic acid). In addition, the scheme that we propose to synthesise glycolaldehyde from ethanol naturally explains the correlation seen by Lefloch et al. (2017) between the abundances of these two species.

DISCUSSION
On the contrary, the abundance of acetic acid and formic acid is predicted to be about one order of magnitude larger than the ones measured and reported in Table 4. We emphasise, however, that the only source in Table 4 for which there is a measurement of the formic acid abundance and an upper limit to that of acetic acid is L1157-B1, which is not a hot corino, so that a more specific modelling is necessary before firmly concluding that there is a problem. Since recent observations of two other iCOMs, acetaldehyde and formamide, in L1157-B1 show that there is segregation in their spatial distribution, we postpone such a modelling to a dedicated forthcoming article.
The other Table 4 source with an estimate of the acetic acid abundance and an upper limit to the formic acid one is IRAS16293-2422. In this case, the discrepancy between the model predictions and the observations might suggest that important routes of destruction of the two species are missing in our network. Since we carefully checked all the "usual" ion-neutral reactions of destruction (i.e. with HCO + , H + , H + 3 , H 3 O + and He + ), it is possible that major sinks are due to missing reactions involving abundant radicals. These reactions might possibly lead to an even higher degree of molecular complexity following a scheme similar to the one proposed here.

CONCLUSIONS
We have presented a new scheme for the synthesis of glycolaldehyde, acetic acid and formic acid from reactions involving ethanol as an ancestor species. The ini-  Table 4. Abundances, with respect to H-atoms, of ethanol, glycolaldehyde, acetic acid and formic acid towards three hot corinos, NGC1333 IRAS4A and IRAS2A , and IRAS16293-2422(Jorgensen et al. 2016and Jaber et al. 2014, and the shock site L1157-B1 (Lefloch et al. 2017).
tiating reactions, with H-abstraction from ethanol leading to two different reactive radicals, have been characterized in laboratory experiments (see Section 3), even though further experimental work at the relevant temperature is mandatory to determine the product branching ratios. The subsequent reactions were considered here for the first time. As there was no information on the complete chemical scheme, we have performed dedicated electronic structure and kinetics calculations to derive the rate coefficients and product branching ratios (Section 4). The rate coefficients of the new reactions were inserted in an updated chemical network and we ran several models to predict abundances of glycolaldehyde, acetic acid and formic acid as a function of the abundance of ethanol. The predictions compare extremely well with the measured abundance of glycolaldehyde in solar type hot corinos and shock sites, both in terms of absolute abundance and in reproducing the correlation between the ethanol and glycolaldehyde abundances observed by Lefloch et al. (2017). Needless to say, more observations towards hot corinos are mandatory to assess the robustness of our new network of reactions leading to glycolaldehyde. The new observations and the detailed models of those sources will be able to discriminate whether the grain or gas-phase chemistry or a combination of the two are mainly responsible for glycolaldehyde formation.
On the contrary, acetic acid and formic acid are predicted to be about ten times more abundant than the extremely sparse detections so far available towards hot corinos and shock sites (only one in each case). This might point to a lack of important routes of destruction of these two molecules in our network, possibly via reactions involving radicals. Nonetheless, since observations are published towards only two sources and the model presented here contain a very generic description of hot corinos conditions, more observations and sourcededicated modelling are necessary to confirm this discrepancy.
A more general conclusion is that the new gas-phase scheme suggested in this article increases the number of studies that show the important and previously overlooked role of neutral gas-phase chemistry in the synthesis of iCOMs. First, since the detection of iCOMs in cold prestellar objects (Bacmann et al. 2012, Cernicharo et al. 2012, Vastel et al. 2014, Jimenez-Serra et al. 2016 it has been clear that gas-phase reactions have to be relatively efficient at 10 K (Vasyunin & Herbst 2013, Vasyunin et al. 2017) as the so-called warm-up phase necessary in pure grain-surface models does not take place in those objects. Second, new studies challenge the exclusive role of grain-surface chemistry in the synthesis of iCOMs also in the warm regions like hot corinos and shock sites , Taquet et al. 2016. The present study adds up new evidence that gas-phase chemistry in the iCOMs synthesis has been overlooked, since important reactions are missing in the current astrochemistry databases.
However, the word "end" cannot be written yet, as more studies are necessary. On the one hand, observations are too scarce to draw firm conclusions and, on the other hand, more theoretical and experimental studies are needed to complete the gas-phase networks. New experimental results on neutral-neutral reactions at low T, previously disregarded because of the presence of an entrance barrier, promise to boost the role of gas-phase reactions involving radicals (Potapov et al. 2017).
NB acknowledges the financial support from the Université Grenoble Alpes and the Observatoire de Grenoble.
This work has also been supported by MIUR "PRIN 2015" funds, project "STARS in the CAOS (Simulation Tools for Astrochemical Reactivity and Spectroscopy in the Cyberinfrastructure for Astrochemical Organic Species)", Grant Number 2015F59J3R.