Entrapment of CO in CO2 ice

TL;DR

This study investigates whether CO2 ice can trap CO molecules effectively, finding that CO2 ice traps CO more efficiently than water ice under certain conditions, which impacts models of volatile distribution in planet formation.

Contribution

It provides experimental evidence that CO2 ice can trap CO more efficiently than water ice, influencing our understanding of volatile partitioning in planet-forming disks.

Findings

CO2 ice traps 40-60% of CO molecules during desorption experiments.

CO2 ice traps CO more efficiently than H2O ice up to ~70K.

Entrapment efficiency increases with ice thickness and CO dilution.

Abstract

Planet atmosphere and hydrosphere compositions are fundamentally set by accretion of volatiles, and therefore by the division of volatiles between gas and solids in planet-forming disks. For hyper-volatiles such as CO, this division is regulated by a combination of binding energies, and by the ability of other ice components to entrap. Water ice is known for its ability to trap CO and other volatile species. In this study we explore whether another common interstellar and cometary ice component, CO2, is able to trap CO as well. We measure entrapment of CO molecules in CO2 ice through temperature programmed desorption (TPD) experiments on CO2:CO ice mixtures. We find that CO2 ice traps CO with a typical efficiency of 40-60% of the initially deposited CO molecules for a range of ice thicknesses between 7 and 50ML, and ice mixture ratios between 1:1 and 9:1. The entrapment efficiency increases with ice thickness and CO dilution. We also run analogous H2O:CO experiments and find that under comparable experimental conditions CO2 ice entraps CO more efficiently than H2O ice up to the onset of CO2 desorption at ~70K. We speculate that this may be due to different ice restructuring dynamics in H2O and CO2 ices around the CO desorption temperature. Importantly, the ability of CO2 to entrap CO may change the expected division between gas and solids for CO and other hyper-volatiles exterior to the CO2 snowline during planet formation.

Figures

Tables

Table 1. Table 1 Integration ranges and band strengths (A) of CO, CO 2 and H 2 O ices.

Species	IR Band	Band pos. (cm-1)	A (cm)
CO	C-O str.	2120-2170	1.4 x 10-17
CO2	C-O str.	2320-2380	1.3 x 10-16
H2O	O-H str.	3000-3600	2.2 x 10-16

Table 2. Table 2 List of Experiments for CO 2 :CO ices

No.	CO2	CO	Initial CO2:CO	CO${}_{}{}^{\ast }{}_{\mathrm{Trapped}}{}^{}$	CO${}_{\mathrm{Trapped}}{}^{\ast }/$	Trapped CO2:CO∗∗
	(ML)	(ML)	mixture at T=20 K	(ML)	COInitial (%)	mixture at T=70 K
1	6.7	6.6	1:1	0.94 / 0.5	14% / 8%	7:1
2	7.1	4.6	2:1	1.1 / 0.93	24% / 20%	7:1
3	21	13	2:1	4.4 / 2.6	32% / 20%	5:1
4	23	15	2:1	3.2 / 3.6	21% / 24%	7:1
5	30	13	2:1	5.0 / 3.8	40% / 31%	6:1
6	31	13	2:1	5.1 / 3.9	40% / 30%	6:1
7	32	10	3:1	4.6 / 4.5	45% / 43%	7:1
8	53	20	3:1	9.3 / 8.2	46% / 41%	6:1
9	5.3	1.3	4:1	0.57 / 0.69	44% / 53%	11:1
10	8.2	2.3	4:1	1.1 / 1.6	46% / 66%	9:1
11	24	6.2	4:1	3.6 / 4.0	55% / 65%	7:1
12	31	7.1	4:1	4.0 / 4.0	57% / 56%	8:1
13	37	9.5	4:1	4.8 / 4.9	50% / 51%	8:1
14	42	11	4:1	6.5 / 6.2	60% / 57%	6:1
15	16	3.1	5:1	1.5 / 2.1	49% / 67%	11:1
16	29	5.5	5:1	2.8 / 3.7	51% / 67%	10:1
17	36	4.8	7:1	2.8 / 3. 8	59% / 80%	12:1
18	10	1.3	8:1	0.60 / 0.94	53% / 74%	17:1
19	5.7	0.66	9:1	0.27 / 0.5	43% / 71%	21:1
20	30	3.4	9:1	2.1 / 3.0	60% / 87%	14:1

Table 3. Table 3 List of Experiments for H 2 O:CO ices

No.	H2O	CO	Initial H2O:CO	CO${}_{}{}^{\ast }{}_{\mathrm{Trapped}}{}^{}$	CO${}_{\mathrm{Trapped}}{}^{\ast }/$	Final H2O:CO∗∗
	(ML)	(ML)	at mixture at T=20 K	(ML)	$C{O}_{\mathrm{Initial}}$	mixture at T=130 K
1	12	12	1:1	0.35 / 0.68	3% / 6%	35:1
2	6.8	4.1	2:1	0.1 / 0.5	2% / 12%	$>$58:1
3	15	5.9	2:1	0.12 / 0.5	2% / 8%	120:1
4	20	9.9	2:1	0.50 / 1.6	5% / 16%	41:1
5	41	19	2:1	1.3 / 3.1	7% / 17%	31:1
6	61	35	2:1	3.2 / 6.0	9% / 17%	19:1
7	28	11	3:1	1.1 / 1.9	10% / 17%	26:1
8	30	9.2	3:1	1.1 / 2.1	12% / 23%	27:1
9	18	3.3	5:1	0.60 / 1.8	18% / 55%	30:1
11	28	2.8	10:1	0.70 / 1.9	26% / 66%	40:1
10	28	2.7	10:1	0.96 / 1.5	34% / 59%	29:1

Equations

$$N_i=\frac{cos(\theta )\int {\tau }_i(v)dv}{A_i}$$

Full text

1. Introduction

Planets and planetesimals acquire their volatiles through ice and gas accretion in protoplanetary disks. The division of volatiles between icy grain mantles and gas across the disk is a key ingredient for predicting atmosphere and hydrosphere compositions of planets forming at different disk locations (Lewis, 1974; Stevenson & Lunine, 1988; Öberg et al., 2011b). To a first approximation volatiles such as water, CO2 and CO are distributed between ice and gas in disks based on their respective sublimation energies, and the temperature profiles of disks. Disk temperature profiles are mainly set by stellar radiation and by accretion heating (D’Alessio et al., 1999), resulting in a decreasing temperature with distance from the star. This temperature gradient should result in a series of condensation fronts or snowlines. Where a planet forms with respect to these snowlines will determine its initial composition, including the volatile elemental ratio, e.g. C/O/N ratios (Piso et al., 2016; Cridland et al., 2016).

In reality the division of volatiles between solids and gas is more complex than encoded in the snowline concept. Hyper-volatiles such as CO, N2 and Argon can become entrapped in less volatile ice, especially in water ice, and therefore reside in disk solids much closer to the star than their sublimation energies would suggest (e.g. Bar-Nun et al., 1985; Visser et al., 2009; Marty et al., 2017). Such entrapment has been used to explain volatile abundances across the Solar System, including C and N enrichment in Jupiter (Owen et al., 1999), and CO and N2 abundances in comets (Notesco & Bar-Nun, 1996; Notesco et al., 1997; Bar-Nun et al., 2007; Mousis et al., 2012; Lectez et al., 2015; Rubin et al., 2015). Experiments on entrapment kinetics together with cometary hypervolatile abundances have also been used to constrain whether cometary ices are interstellar or formed in the Solar Nebula (Notesco et al., 2003; Mousis et al., 2016). Entrapment of volatile radicals in H2O ice may also be key to grow chemical complexity in interstellar and circumstellar ices (Garrod et al., 2008; Fresneau et al., 2014).

So far, ’entrapment’ has been used to denote any mechanism through which hyper-volatiles are retained in an ice matrix beyond their normal sublimation temperatures. There are at least two distinct entrapment mechanisms, however, ‘mechanical’ entrapment and clathration (e.g. Hersant et al., 2004), which are active under different interstellar and nebular conditions. In short, ’mechanical entrapment’ (from now on, simply entrapment) occurs when hyper-volatiles are trapped in pores in amorphous ice. By contrast, clathrate formation requires crystalline water ice, in which cages are stabilized by the inclusion of the hyper-volatile in lattice structure (e.g. Lunine & Stevenson, 1985). While the latter mechanism requires water ice, the former could in theory work for any amorphous ice with a porous structure.

Even though entrapment could take place in any amorphous ice, the focus has been on water ice, the most abundant ice constituent in the interstellar medium, around low and high-mass protostars (Gibb et al., 2004; Öberg et al., 2011a; Boogert et al., 2015), and in comets (Mumma & Charnley, 2011). Based on numerous experiments, amorphous water ice can indeed trap volatiles (Bar-Nun et al., 1985; Hudson & Donn, 1991; Collings et al., 2003; Loeffler et al., 2006; Fayolle et al., 2011, e.g.), and this may explain why e.g. some comets are rich in CO despite multiple past heating events above the CO sublimation temperature (Mumma & Charnley, 2011).

Water is not the only abundant ice constituent, however. CO2 is typically present at a 20-40% abundance with respect to water in interstellar, circumstellar and cometary ices alike (Boogert et al., 2015; Mumma & Charnley, 2011). In interstellar and circumstellar ices, a majority of the CO2 is mixed in with the water ice, but some is also present in a separate, CO and CO2 dominated ice phase (Pontoppidan et al., 2008). The ice morphology in comets is largely unknown since ices are not observed directly, but for two comets 81P/Wild 2 and 67P/Churyumov-Gerasimenko, there is evidence for (partially) separate water and CO2 ice phases. In the case of 81P/Wild 2, CO2 and water outgassing appeared from different cometary locations, indicative of that some portion of the CO2 is in a separate reservoir (Mumma & Charnley, 2011). During observations of 67P/Churyumov-Gerasimenko, CO2 outgassing was also not well correlated with water outgassing, and indeed often anti-correlated (Hoang et al., 2017; Gasc et al., 2017). While different sublimation temperatures of H2O and CO2 may contribute to this behavior it unlikely explains it completely, since in laboratory experiments substantial amounts of CO2 co-desorb with the water ice if the the two are originally mixed together (e.g. Fayolle et al., 2011). Furthermore, the CO2 outgassing did correlate with the outgassing of CO (Hässig et al., 2015), which raises the possibility that the observed CO was trapped in the CO2 ice.

Based on these clues we have undertaken an experimental study into the entrapment of CO in CO2 ice, starting with amorphous CO2:CO ice mixtures. §2 presents the experimental apparatus and methods. In §3 we first describe the outcome of 20 CO2:CO ice experiments, and how entrapment depends on the initial ice thickness and ice mixing ratio. We then present results from a smaller set of analogous H2O:CO experiments and compare the entrapment efficiencies in H2O and CO2 ice. §4 discusses the experimental trends, possible origins for differences between CO2 and H2O entrapment efficiencies, and implications for volatile compositions across planet forming disks. §5 presents some concluding remarks.

2. Experimental methods

2.1. Experimental set-up

The experimental set-up consists of a 6.5″ ID ultra-high vacuum (UHV) chamber with a base pressure of 4$\times 10^{-9}$ Torr at room temperature. The chamber is designed to investigate thermal processes in interstellar ice analogs. The chamber is evacuated using a turbomolecular pump (Pfeiffer Vacuum Inc. HiPace® 300 with pump speed of 260 l/s for N2) backed by a rotary vane pump (Pfeiffer Adixen 2005 SD Pascal) with an oil mist filter. The chamber pressure is monitored by MKS series 999 Quattro multi-sensor vacuum transducer coupled with a series PDR900 single channel controller which provides a continuous measurement from 10*-10* Torr to atmosphere. The UHV chamber is placed inside the sample compartment of a standard Fourier Transform Infrared (FTIR) spectrometer (Bruker Optics Inc. Vertex 70) with a liquid N2 cooled Mercury Cadmium Telluride (MCT) detector. The IR spectrometer and the UHV chamber are placed on separate solid frames equipped with vibration isolation slides such that the optical path of the IR instrument is aligned with the horizontal axis of the chamber in the transmission geometry. The UHV chamber is separated from the spectrometer through two differentially pumped IR transparent KBr windows on either side. The entire optical path of the IR chamber is continuously purged with dry CO/CO2 free air using a Parker 75-62 purge gas generator.

The UHV chamber houses an IR transparent 2mm thick CsI window (0.75″ clear view) mounted on an optical ring sample holder attached to a closed cycle helium cryostat (Advanced Research Systems (ARS) model DE-204S). The cryostat is integrated with DMX-20B interface which makes use of helium exchange gas to decouple the CsI window from the cold tip of the cryocooler, which prevents almost all vibrations from being transferred to the CsI window. The cryostat expander is also isolated from the sample mount via isolation bellows, which work in conjunction with gas heat exchange interface to avoid direct transmission of vibration to the IR spectrometer as well. The nickel plated copper radiation shield attached to the first stage of the cryostat helps to achieve CsI substrate temperature down to 17K. Note that because the shield is only cooled using the first stage it is too warm for very volatile ices such as CO to freeze out on it, while less volatile ices such as H2O may condense there if an ice deposited through background deposition (not used in the experiments presented here). The substrate temperature is controlled between 17K and 350K with a relative precision of +/-0.1K by a 75W resistive heater and a calibrated Si diode sensor (LS-670B-SD) located on the cryocooler tip near the heater using an external temperature controller (Lakeshore Model 335). A second identical Si diode sensor is mounted directly on the CsI substrate to measure the actual ice temperature. The absolute temperature uncertainty is estimated to 2–3 K.

The differentially pumped rotary seal (Thermionics Vacuum Products, RNN-400) on the cryostat assembly allows 360 degree rotation of the CsI window inside the chamber during experiment. Ices are grown on the crycooled CsI window through directed vapor deposition using a gas dozer comprising a 4.8 mm ID stainless steel tube attached to a VAT variable leak valve and compact z-stage (MDC Vacuum Products, LLC). A stainless steel gasline is used to store gases and prepare gas mixtures of interest and deliver them into the chamber using the gas doser. The gasline is pumped down to pressures $<$5 x 10*-4* Torr using a turbopump station (Pfeiffer Vacuum Inc. HiCube Eco with pump speed of 67 l/s for N2). The pressure in the gasline is monitored using two active capacitance transmitters (Pfeiffer Vacuum Inc.) CMR 361 and CMR 365 from $5\times 10^{-4}$ to 1100 Torr.

2.2. Experimental procedures

The experiments presented in this study are all temperature programmed desorption (TPD) experiments, in which a CO2:CO or H2O:CO ice mixture is heated using a linear heating ramp until desorption is complete. CO entrapment is measured by monitoring the desorption rate of CO during the ice warm-up using the QMS, and by monitoring the ice composition using the FTIR.

Gas mixtures are prepared inside the gasline attached to the chamber within an hour of the experiment. We also ran a few experiments where we let the gas-mixture sit overnight and found no significant difference in the experimental outcome. In the experiments, we used CO2 (99%, Sigma-Aldrich), CO (99%, Sigma-Aldrich) and deionized H2O liquid purified through multiple freeze-pump-thaw cycles.

The gas mixture is deposited onto the cryocooled substrate at 17 K until the desired ice thickness has been achieved. The latter is measured using infrared spectroscopy. In all experiments FTIR spectra are acquired at a spectral resolution of 1 cm*-1* from 4000 to 400 cm*-1*. Each acquired spectra consists of 128 averaged scans and takes $\sim$3 min to complete. Background spectra are recorded before each ice deposition and subtracted from the respective ice spectra.

IR spectral band integrated opacities can be converted into ice column densities using the band strengths in Table 1 from Hudgins et al. (1993) and Gerakines et al. (1995), corrected for ice density in Bouilloud et al. (2015). We operate in transmission mode with the substrate at a 45∘ angle to the beam path. the calculation of ice column density is therefore

[TABLE]

where Ni is the column density (molecule.cm*-2*), cos($\theta$) is the angle between the IR field vector and the ice surface normal taking into account refraction at the vacuum-ice interface, $\int\tau_{i}(v)dv$ is the integrated optical depth of the IR band area, and $A_{i}$ is the band strength of the species. The column density in molecules cm*-2* is converted to monolayers of ice (ML) by using a standard 1015 molecules cm*-2* conversion factor. We use the Eq. 1 both to calculate the initial ice matrix thickness (CO2 or water) and the matrix:CO mixing ratio. The formal uncertainty in these calculations, based on the spectral rms and line widths, are $\sim$0.1 ML, but due to uncertainties in baseline subtractions we do not attempt to quantify ice thicknesses below 0.5 ML. The band strengths have reported uncertainties of 5–20%, but this is for a specific temperature and ice mixture environment. In the case of CO, changes on the order of 10–30% have been reported both when considering CO in different ice matrices at a single temperature, and for the same ice observed at different temperatures (e.g. D’Hendecourt & Allamandola, 1986; Schmitt et al., 1989; Gerakines et al., 1995). The reported ice mixture ratios and thicknesses are therefore only assumed to be accurate within 30%.

Following ice deposition, the ice mixtures are heated with a constant rate of 1 K/min until complete desorption of the ice mixture in a so called temperature programmed desorption experiment (TPD). During the TPD, the ice desorption rate is measured every 1-2 seconds using the QMS, and IR spectra of the ice are acquired every 5 min. The shape and locations of TPD QMS curves encode information about the ice binding environment and structure, but in these experiments we only use the TPD curves to measure entrapment fractions based on the TPD characteristic that the integrated areas of a TPD curve between two time stamps is proportional to the number of molecules desorbed during that time.

The TPD curves are recorded as ion counts by the QMS, and this is converted into desorbing ML per K, by calibrating the total integrated TPD curve to the initial ice quantity measured by the FTIR. The smallest entrapment CO quantities we report with the QMS are 0.1 ML, since for smaller quantities the TPD baseline subtraction is uncertain. One complication that needs to be accounted for is that CO2 partially fractionates into CO in the QMS. According to the NIST database, the strength of the signal is $\sim$10% compared to the main m/z=44 signal. The exact electron impact energy of those measurements is not listed, however, and we also ran local tests, which were consistent with the NIST value. In all CO2:CO TPD experiments 10% of the CO2 signal is therefore subtracted from the CO signal before any analysis.

3. Results

3.1. CO entrapment measurements

During TPD experiments we expect, based on previous experiments on H2O:CO ice mixtures (e.g. Collings et al., 2003; Fayolle et al., 2011), CO to desorb at two or more distinct temperatures corresponding to CO molecules escaping around the ’normal’ CO desorption temperature of $\sim$30 K, and CO molecules that are trapped in the ice matrix. The latter CO may either be retained in the ice until the ice matrix desorbs around 80 and 150 K, for CO2 and H2O ice mixtures respectively, or escape during ice matrix restructuring during phase changes (e.g. Bar-Nun et al., 1985; Hudson & Donn, 1991; Collings et al., 2003, 2004). Figure 1 (left panel) shows that in the case of CO:CO2 ice mixtures, most of the CO does indeed desorb either around 30 K, or together with CO2 desorption. In addition we observe a low level of CO desorption around 55 K, which we attribute to out-gassing of CO from the CO2 ice matrix during CO2 ice restructuring; a change in CO2 ice spectral features has been observed around 60 K, attributed to the completion of CO2 ice crystallization (He & Vidali, 2018).

The fraction of trapped CO is obtained by integrating the CO QMS signal for temperatures $>$65 K, the earliest onset of CO2 desorption in the experiments, and dividing it by the QMS signal integrated over the entire temperature range. In the experiment shown in Figure 1 the measured CO entrapment fraction is 51%.

The QMS measurements are complemented by FTIR spectral time series that probe the CO ice quantity (Fig. 1, middle and right panels). The infrared spectra are converted to CO column densities in units of ML as described in §2.2. Converting CO spectra into a CO quantity is complicated by a possible dependence of the CO ice band strength on ice morphology and composition, both which are expected to change during the warm-up ramp due to CO and ice matrix structural changes (e.g. Schmitt et al., 1989; Isokoski et al., 2013; He & Vidali, 2018). Figure 1 shows that indeed, while the QMS and FTIR CO measurements are broadly in agreement, there are quantitative differences. First, while there is a decrease in the CO ice IR feature strength at $\sim$30 K, corresponding to where one of the two major CO desorption peaks observed in Fig. 1 (left panel), the decrease in the IR is comparatively small. Second there is a sharper than expected decrease in the CO ice feature around 60 K. Third, for some CO2:CO and all H2O:CO entrapment experiments, the entrapment measured by the FTIR is considerably higher than those measured with the QMS; we estimate the FTIR CO entrapment fraction by dividing the CO ice abundance at 65 K (100 K in water ices) with the initial CO ice abundance.

There are at least three possible explanations for this discrepancy: ice structure and temperature-dependent CO band strengths, diffusive ice build-up on the backside of the substrate, and uneven ice deposition across the frontside of the substrate. The first two would affect the IR data, while the latter may result in an underestimate of entrapment using the QMS data.

Band strengths are sensitive to ice structure and temperature. Schmitt et al. (1989) found a substantial decrease in the CO band strength when heating up a previously annealed H2O:CO mixture. This may account for the larger than expected decrease in CO intensity between 50 and 70 K in Fig. 1. We also suspect that the band strength may increase around 30 K due to ice structure changes, which would explain the surprisingly small decrease in CO intensity during the first CO desorption peak, but this has not yet been tested experimentally.

While diffusive deposition of ice on the substrate backside could theoretically contribute to the IR/QMS discrepancy, we think it can be ruled out based on standard gas kinetic theory and the pumping rate and geometry of our system. We estimate that the ratio between backside and frontside deposition should be at most 5%, i.e. very small. In set-ups where this is an important factor, IR measurements would underpredict entrapment since some of the measurements would be based on the thinner backside ice, where entrapment is likely less efficient.

Third, ice deposition on the frontside of the sample is expected to present a thickness gradient from the substrate center, since the gas is deposited through a narrow tube close to the substrate. This entails that while the IR probes an ice of almost constant thickness, the QMS will also measure entrapment fractions in the thinner ices deposited further from the substrate center on the frontside of the substrate, and on the sample holder. If this effect is important, it will result in a systematic underprediction of entrapment when using the QMS data.

With these uncertainties in mind, we record entrapment fractions using both methods. When exploring how entrapment fractions depend on ice thickness and compositions we mainly rely on the QMS data, which has a smaller experimental scatter. Since the systematic errors may be different for different ice matrix species we use both IR and QMS data when comparing entrapment in water and CO2 ices.

3.2. CO2:CO Experiments

Informed by observed interstellar ices, the CO2:CO experiments include CO2 ice thicknesses between 7 and 53 ML, and CO2:CO mixing ratios between 1:1 and 9:1 (Pontoppidan et al., 2008). To disentangle the effects of ice thickness and composition, most ice mixtures were run with a CO2 ice thickness of $\sim$30–35 ML, and one ice mixture (4:1) includes an order of magnitude range in ice thicknesses. Table 2 lists all CO2:CO entrapment experiments, organized by CO2 ice thickness, in ML, then by CO ice thickness for each ice mixing ratio.

Figure 2 presents the CO TPD spectra listed in Table 2 in the same order. All TPD spectra show two major CO desorption peaks corresponding to CO desorbing from the ice surface, and co-desorption of entrapped CO at the CO2 desorption temperature, and generally also some desorption at intermediate temperatures as discussed in the previous section. In most experiments the intermediate peak occurs above 50 K, but in the thinnest and only 1:1 mixing ratio experiment it occurs already at 50 K, possibly because of a different ice re-structuring dynamics in this highly concentrated ice. The relative size of the two major CO desorption peaks vary in the different experiments; in TPD experiments with similar ice mixing ratios, the ratio of the entrapment peak to the CO desorption peak increases with ice thickness, and for experiments with similar CO2 ice thickness, the ratio increases with decreasing CO concentration. The relative size of the intermediate peak decreases with ice thickness and with dilution.

CO entrapment in CO2 ices is quantified in Table 2 using both the QMS and FTIR approaches. Using the QMS data, the CO trapping fraction varies between 14 and 60% with respect to the initial CO content. For the majority of CO2:CO experiments, the entrapment efficiency is $>$40%, and for all but 3 experiments it is $>$30%. The only experiments with a low ($<$30%) entrapment efficiency are both highly concentrated in CO (CO2:CO mixing ratios of $<$2), and thin ($<$25 ML of CO2).

Table 2 can also be used to estimate the entrapment measurements error. One of the experiments is exactly replicated (Exp 5 and 6), and the resulting difference is 2% in the quantity of trapped CO. If Exp. 7 is also considered a duplicate (it almost is), the error increases to 4% for the QMS measurements and 7% for the IR measurements. The QMS uncertainty is added in quadrature to 0.1 ML, the fundamental TPD uncertainty, to all QMS entrapment measures in the subsequent analysis.

Figure 3, upper panel illustrates how the CO entrapment efficiency depends on ice thickness for a single ice mixing ratio (4:1). When considering the quantity of trapped CO, there is a monotonic increase with CO2 ice thickness between 7 and 50 ML. However, the entrapment efficiency, in % of the original CO, levels out above an initial CO2 ice thickness of 20 ML.

The mixing ratio also influences CO entrapment, which is exemplified in the lower panel of Figure 3 for CO2:CO ices with initial CO2 ice thickness of 30–35 ML. The quantity of CO trapped (black line) decreases with increasing mixing ratio. This is a trivial result, however, since if the ice thickness is constant and the mixing ratio increases there is simply less CO to start with in the ice. By contrast, the fraction of CO trapped increases with the mixing ratio between 2:1 and 4:1, before leveling out.

Figure 4 shows the relationship between entrapment and CO2 ice thickness for all different mixing ratios between 1:1 and 9:1 (Exp. 4 is excluded since it is right in between 1:1 and 2:1). For convenience the 7:1, 8:1 and 9:1 experiments are grouped together. The trends observed for 4:1 ice mixtures and different ice mixtures for CO2 ice thicknesses of 30–35 ML appear to hold for the full set of experiments. That is, the quantity of trapped CO increases monotonically with CO2 ice thickness, and with increasing CO concentration for a given ice thickness. The percentage of trapped CO increases with ice thickness and with decreasing CO concentration when the ice thickness is $<$20 ML and the CO2:CO ice mixing ratio is $<$4:1. For thicker and more dilute ice mixtures, the CO entrapment efficiency levels off.

In addition to entrapment efficiencies, we can also determine how the CO2:CO ice mixing ratio for trapped CO, i.e. how many CO2 molecules are present in the ice per trapped CO molecule at the onset of CO2 desorption, depends on the initial ice thickness and mixing ratio. Inspecting Table 2, the entrapment mixing ratio appears to increase with increasing initial mixing ratio for similar ice thicknesses, while the impact of ice thickness is less clear. This is illustrated in Fig. 5, which shows a strong relationship between the initial and final mixing ratios, such that the most concentrated final CO2 ices are achieved in the experiments with the initially most concentrated ices. The thinnest ices (CO${}_{2}<10$ ML) show systematically higher CO2:CO entrapment mixing ratios, indicative of less efficient entrapment in the thinnest ices, but there is no clear trend with thickness above 10 ML. The maximum trapped CO2:CO mixing ratio observed in the experiments is 5:1, i.e. at least five CO2 molecules is needed to trap one CO molecule in ices up to $\sim$30 ML in thickness.

3.3. H2O:CO Experiments

Table 3 lists experiments on CO entrapment in H2O:CO ices, analogous to Table 2 for CO2:CO experiments. H2O:CO TPD experiments were run for a similar range of ice thicknesses and mixing ratios as CO2:CO. The number of H2O:CO experiments is, however, smaller, since their main purpose is to serve as a comparison set to the CO2:CO experiments. Figure 6 shows CO desorption in the H2O:CO TPD experiments in the same order as listed in Table 3. Similar to the CO2:CO experiments, there are two desorption peaks, around 30–40 K, and around 150 K. The second peak slightly precedes the H2O desorption peak, and based on previous experiments it is due to CO outgassing during H2O ice crystallization (e.g. Collings et al., 2003). By contrast to the CO2:CO experiments, the first peaks always dominates, indicative of that only a small fraction of the initial CO is trapped.

The CO entrapment fractions in the H2O:CO experiments are listed in Table 3 and illustrated in Fig. 8. Compared to the CO2 experiments there is a larger discrepancy between QMS and FTIR results; the FTIR time series suggests on average a factor of 2 higher entrapment fraction. Similar to the CO2:CO experiments, the spectra of CO in cold and warm H2O ice are distinctly different (Fig. 7), suggesting that the CO band strengths may also be different in the two temperature regimes, but as discussed above, the QMS data could also be underestimating the entrapment fraction. According to the QMS measurements the CO entrapment efficiency is $<$10% in a majority of experiments. While the entrapment is higher when basing the analysis of the IR measurements, the entrapment efficiency is still low, $<$25%, in all but the three most dilute ices. This can be compared to the CO2:CO experiments where only a small fraction of thin and highly concentrated ice mixtures resulted in a similarly low CO entrapment efficiency. The CO entrapment trends are similar to those observed in CO2:CO ices, i.e. the entrapment efficiency increases with ice thickness and decreasing CO concentration, which is also in qualitative agreement with previous studies on CO:H2O ices (Hudson & Donn, 1991; Fayolle et al., 2011).

3.4. CO entrapment in CO2 vs. H2O ices

Figure 9 shows a direct comparison of CO entrapment efficiencies in CO2 and H2O ices. Five of our CO2:CO ice experiments are well matched to H2O:CO experiments in terms of mixing ratio and ice matrix thickness. These are experiments number 2, 3, 7, 15 and 20 in Table 2, and 2, 4, 7, 9 and 10 in Table 3. These experiments are compared pairwise in Fig. 9, though note that CO is considered trapped if it is present above 65 K in CO2 ice and above 100 K in H2O ice. Solid bars compare entrapment using the QMS data, and in all cases, substantially more CO is retained in the CO2 ices compared to the analogous H2O ices. The difference is the largest for the thinnest and most concentrated ices, where CO2 traps up to an order of magnitude more of the initial CO. For thicker and more dilute ices the difference is smaller, but CO2 still appears to be $>2\times$ more efficient than H2O in retaining CO. The decreasing gap between CO2 and H2O entrapment efficiencies for more dilute ices, indicates that the efficiency of H2O to entrap hyper-volatiles is more sensitive to specific ice characteristics compared to CO2. We also compare entrapment using IR time series (hatched bars), and while CO2 is still the more efficient entrapper, the difference between water and CO2 is smaller: CO2 traps an extra 20-250% CO compared to water in matched experiments.

4. Discussion

In the previous section, we showed that CO is efficiently trapped in CO2 following warm-up of CO:CO2 ice mixtures from 20 K to CO2 desorption. Surprisingly, this entrapment efficiency is higher than in analogous H2O:CO ice experiments. Among the CO2:CO ice experiments the entrapment efficiency increases with ice thickness and dilution, but only up to a point. The origin and implications of these trends are discussed in §4.1. Possible explanations for the higher entrapment efficiency of CO2 compared to H2O are presented in §4.2, and the astrophysical implications for the discovered CO2 entrapment are introduced in §4.3.

4.1. CO entrapment in CO2 ice

The observed CO2 entrapment characteristics and trends reported in §3.2 encode information on under which conditions CO2 ice can retain mixed in CO molecules, and on the entrapment mechanism. First, the entrapment efficiency increases with ice thickness, but only for thin ices, $<$10–30 ML. For thicker ices, the entrapment efficiency is constant. This behavior implies that the entrapment efficiency decreases towards the ice surface in the upper ice layers. Deeper into the ice, CO2 appears to retain a constant fraction of the initially mixed in CO. The CO2 ice thickness required to reach this constant efficiency increases with CO concentration, indicating that CO out-gassing is possible from deeper into the ice in more concentrated ices.

Regardless of ice thickness, there seems to be a minimum required number CO2 molecules per trapped CO molecule. Based on the observed entrapped CO2:CO mixing ratios this number is approximate 5 CO2 molecules per trapped CO molecule. This can be compared e.g. to e.g. studies of H2O clathration, where a similar proportionality is required, suggesting that the two entrapment mechanisms can result in similar number of hyper-volatiles in same-sized ice matrices (e.g. Lunine & Stevenson, 1985).

A second set of constraints on the entrapment mechanics comes from the TPD spectra and FTIR time series. Both show evidence for substantial CO loss in between the CO and CO2 desorption temperatures. This region coincides with a thickness-dependent completion of CO2 crystallization (He & Vidali, 2018). This outgassing of CO is reminiscent of the previously observed outgassing of CO from H2O during H2O crystallization (Bar-Nun et al., 1985; Collings et al., 2003). The results of this phase change is seen both in CO and CO2 spectral shapes. In Fig. 1, the CO band becomes more structured between 50 and 75 K, indicative of that the the ice environment within which CO resides is changing. Concurrently, there is a clear shift and narrowing of the CO2 ice band (Fig. 10).

Finally, spectroscopy of the CO2 ice feature during ice warm-up (Fig. 10) shows that CO2 indeed has undergone some restructuring already by 30 K, similar to what was observed for pure CO2 ice (He & Vidali, 2018). If CO is trapped through pore collapse, analogous to what has been proposed for amorphous water ice (Collings et al., 2003), then the spectroscopy suggests that CO2 mobility and therefore CO2 pore collapse may begin prior to the onset of CO desorption, which may help to explain why it can trap CO molecules so efficiently. It thus appears that CO2 ice restructuring holds a key to both CO retention and CO escape from CO2 ices.

The scenario emerging from the above constraints is the following: At $\sim$30 K, CO becomes volatile in the ice and begins to escape from the CO2 ice surface and pores. Around the same temperature CO2 becomes mobile and begins to crystallize, trapping CO with an increasing efficiency with ice depth. Between 40 and 60 K, CO2 crystallization completes and during this process some of the remaining CO escapes as the CO2 ice restructures, opening some previously closed pores. The finally remaining CO, which is at most at a 1:5 mixing ratio with CO2 in thin ices, stays trapped until the onset of CO2 desorption between 65 and 75 K.

4.2. H2O vs CO2 entrapment

The most surprising outcome of this study is that prior to the onset of CO2 desorption, pure CO2 traps CO molecules more efficiently than comparable H2O ices. This result may be due to differences in the CO2 and H2O ice binding environments, structures, or dynamics. We explore the possible contributions from each in this sub-section.

The CO binding energies to CO2 and H2O molecules in ice have been constrained through TPD experiments in which sub-monolayers of CO are deposited on compact, thermally annealed CO2 and H2O ices. The key result is that the desorption temperatures and therefore binding energies are the same for CO on CO2 and CO on H2O ice within experimental uncertainties (Fayolle et al., 2016; Cooke et al., 2018). By contrast, the CO desorption energy for CO deposited on a porous amorphous H2O ice, constructed at 15 K through direct vapor deposition, is substantially higher than for CO deposited on an analogous CO2 ice (Cooke et al., 2018). This implies that water ice deposited at $\sim$10–15 K can host more CO in strongly bound sites (e.g. in micropores) than CO2 ice deposited at the same temperature. This suggests that low-temperature H2O ice is more porous than low-temperature CO2 ice. Such a porous structure may be detrimental to CO retention in mixed ices if pore collapse is slow compared to CO desorption, but may be beneficial to capture CO into the ice in the first place.

To test whether H2O ice is more efficient than CO2 at entrapping CO molecules not already mixed into the ice, we performed two supplemental, layered ice experiments. In these experiments $\sim$3 ML of CO is deposited on top of $\sim$50 ML of CO2 or H2O ice. The substrate temperature was kept at 15 K during all depositions. Figure 11 shows that also in this scenario CO2 traps substantially more of the deposited CO compared to the H2O ice, similar to in the ice mixture experiments. Previous experiments on water ice have shown that the entrapment efficiency increases with ice porosity (Ayotte et al., 2001). This comparison suggests that the difference between CO2 and H2O ice entrapment efficiencies is not mainly determined by different initial surface pore distributions.

Finally we consider the different ice restructuring processes in H2O and CO2 ices as a possible cause of the different entrapment efficiencies. Both water and CO2 undergo phase changes between the onset of CO desorption around 30 K, and 70 K. In the case of water this is a phase change between low-density amorphous water and high-density amorphous water (e.g. Collings et al., 2003), while in the case of CO2 it is a drawn-out phase change from amorphous CO2 to crystalline CO2 ice (Isokoski et al., 2013; He & Vidali, 2018). The latter transformation may be more efficient in trapping CO in the ice. CO2 ice restructuring may also begin earlier than the water phase change due to higher CO2 ice mobility. Indeed Escribano et al. (2013) and He & Vidali (2018) show that some CO2 restructuring begins already at 20 K, and CO2 may thus begin to trap CO before the onset of CO desorption.

4.3. Astrophysical implications

During planet formation, the distribution of abundant volatiles between the solid and gas phases regulates the final compositions of planetary cores, hydrospheres and atmospheres. Entrapment of hyper-volatiles such as CO and N2, in less volatile ice matrices may change this distribution substantially, maintaining e.g. CO far interior to its expected condensation front or snowline. In this paper we have shown that water ice is not the only common interstellar ice that can entrap molecules. CO2, generally the second most abundant ice constituent (Öberg et al., 2011a; Boogert et al., 2015), may also trap CO during planet formation, and at a higher efficiency than H2O in thin ices, as long as the ice is too cold for CO2 sublimation. Since CO2 is an abundant ice constituent, the identified difference in entrapment efficiency of $\sim$2 implies that H2O and CO2 ice may trap comparable quantities of CO in the ISM and in planet forming disks if H2O and CO2 reside in different ice phases. This is a big ’if’ since in star forming regions most CO2 is mixed in with water ice. As already introduced in §1 there are several reasons why the case may be different in disks. First, the two comets studied in detail seem to present H2O and CO2 sublimations from different comet locations (Mumma & Charnley, 2011; Hoang et al., 2017; Gasc et al., 2017). While some of this difference can be explained by differences in sublimation temperatures between H2O and CO2, a substantial amount of CO2 would be expected to co-desorb with H2O if they were perfectly mixed together (Fayolle et al., 2011). Second there is theoretical evidence for substantial conversion of CO into CO2 in protoplanetary disks based on astrochemical models (Schwarz et al., 2018) and low CO/H2 ratios (Bergin et al., 2013). Such a conversion on the surfaces of grains would naturally result in a CO2:CO ice phase separate from that of H2O.

Assuming then that all H2O and CO2 ice are available for CO entrapment, we can ask how much CO could be retained in ice grains in disks interior to the nominal CO snowline. Our study suggests that a substantial fraction could be present in CO2 ice alone between the CO and CO2 snowlines. In star-forming clouds, CO2 ice abundances are as high as $2.6\times 10^{-5}$ per $n_{\rm H}$ (Boogert et al., 2015). Adopting our maximum observed entrapment CO2:CO mixing ratio of 5:1, the maximum CO abundance that could be trapped in CO2 ice is $5.2\times 10^{-6}$ per $n_{\rm H}$. This is comparable to estimates of CO abundances in some protoplanetary disks (Bergin et al., 2013; Schwarz et al., 2016; Cleeves et al., 2016; Miotello et al., 2017; Long et al., 2017). It is important to note, however, that not all or even not most CO2 may be present as pure ice, and future studies should explore CO entrapment in H2O:CO2 mixed ices, similarly to those commonly identified in the ISM.

Our experiments only probes entrapment of CO. Whether CO2 also efficiently entraps N2, CH4, O2 and noble gases remains to be explored. The proposed entrapment mechanism due to pore collapse around 30 K should be species agnostic, but it is possible that less volatile molecules than CO, e.g. N2, begins to desorb prior to the onset of pore collapse. If this is the case, selective CO2 entrapment may provide an alternative explanation for the low N2/CO ratio observed in most comets, most recently in 67P (Rubin et al., 2015), than preferential CO clathration in water ice (Mousis et al., 2012; Lectez et al., 2015).

5. Conclusion

Entrapment of hyper-volatiles in less volatile ice matrices changes the gas-phase and ice-phase molecular and elemental compositions during planet formation. In this experimental study we show that CO2 ice is a potential trapping agent in the ISM and in planet forming disks. CO2 ice surprisingly traps CO more efficiently than analogous water ices. This may be explained by an earlier onset of pore collapse in CO2 ice compared to H2O ice, trapping a majority of CO molecules prior to the onset of CO desorption. CO2 entrapment efficiencies are 40–60% for the large majority of explored ice compositions and thicknesses, indicating that substantial quantities of CO may be trapped in CO2 ice. This discovery increases the theoretical amount of CO that can be retained in ice in between the CO2 and CO snowlines in disks. It may also help to explain the correlation of CO and CO2 outgassing from 67P.

AS is grateful to Jennifer Bergner for assistance with constructing data analysis scripts. KIÖ acknowledges funding from the Simons foundation through the Simons Collaboration on Origins of Life (SCOL) Investigator Award (#321183)