Binding energies and sticking coefficients of H₂ on crystalline and amorphous CO ice

Abstract

Context. Molecular hydrogen (H2) is the most abundant interstellar molecule and plays an important role in the chemistry and physics of the interstellar medium. The interaction of H2 with interstellar ices is relevant for several processes (e.g., nuclear spin conversion and chemical reactions on the surface of the ice). To model surface processes, quantities such as binding energies and sticking coefficients are required. Aims. We provide sticking coefficients and binding energies for the H2/CO system. These data are absent in the literature so far and could help modelers and experimentalists to draw conclusions on the H2/CO interaction in cold molecular clouds. Methods. Ab initio molecular dynamics simulations, in combination with neural network potentials, were employed in our simulations. Atomistic neural networks were trained against density functional theory calculations on model systems. We sampled a wide range of H2 internal energies and three surface temperatures. Results. Our results show that the binding energy for the H2/CO system is low on average, − 157 K for amorphous CO and −266 K for crystalline CO. This carries several implications for the rest of the work. H2 binding to crystalline CO is stronger by 109 K than to amorphous CO, while amorphous CO shows a wider H2 binding energy distribution. Sticking coefficients are never unity and vary strongly with surface temperature, but less so with ice phase, with values between 0.95 and 0.17. With the values of this study, between 17 and 25% of a beam of H2 molecules at room temperature would stick to the surface, depending on the temperature of the surface and the ice phase. Residence times vary by several orders of magnitude between crystalline and amorphous CO, with the latter showing residence times on the order of seconds at 5 K. H2 may diffuse before desorption in amorphous ices, which might help to accommodate it in deeper binding sites. Conclusions. Based on our results, a significant fraction of H2 molecules will stick on CO ice under experimental conditions, even more so under the harsh conditions of prestellar cores. However, with the low H2–CO binding energies, residence times of H2 on CO ice before desorption are too short to consider a significant population of H2 molecules on pure CO ices. Diffusion is possible in a time window before desorption, which might help accommodate H2 on deeper binding sites, which would increase residence times on the surface.

Floating in Space: How to Treat the Weak Interaction between CO Molecules in Interstellar Ices

“…Overall, the degree of agreement among all DFT methods including dispersion and CCSD­(T) is satisfactory, confirming previous calculations. ,, We observe that B3LYP without dispersion correction yields results that are far from the reference value, while both D4 and NL corrections bring the values much closer to the reference. In general, NL-corrected functionals perform only slightly better compared to those with a D4 correction.…”

“…Computational studies on CO binding energies have been sparse, because finding an appropriate balance between accuracy and computational cost is very challenging for this weakly bound system. , High­(er) levels of ( ab initio ) theory cannot describe systems large enough to mimic amorphous systems. Thus, to allow larger system sizes, other techniques, such as force-field-based classical molecular dynamics, , continuous time random-walk Monte Carlo simulations, and kinetic Monte Carlo simulations , have been applied in recent years.…”

Predicting binding energies of astrochemically relevant molecules via machine learning

“…ML interatomic potentials are another type of ML model, which can be directly employed as a lowcost alternative to quantum chemical calculations for investigating, for example, molecular reactivity, adsorption, and diffusion on dust grains (e.g. Mazo-Sevillano et al 2021;Molpeceres, G. et al 2021;Zaverkin et al 2021). In surface science and catalysis, ML has also been used extensively to identify predictive models for BEs (e.g.…”

Exploring the limits of machine-learned potentials for chemically complex multicomponent systems

“…36 GM-NNs, on the other hand, have so far been developed for a different class of chemical structures. For instance, they were used to study the dynamics of N and H 2 on water and CO ice surfaces, respectively, [37][38][39] and the dynamic behaviour of magnetic anisotropy tensors of [Co(N 2 S 2 O 4 C 8 H 10 ) 2 ] 2− , [Fe(tpa) Ph ] − , and [Ni(HIM 2 -py) 2 NO 3 ] + molecular crystals. 40 However, they have not been tested for metallic, let alone a complex HEA system.…”