Modelling clusters of complex organic molecules

Oxygenated organic molecules (OOMs) form in the atmosphere by oxidation of volatile organic compounds from both natural and anthropogenic sources. Highly oxygenated organic molecules are likely to take part in new particle formation, but it is unclear to what extent they can form particles without the involvement of inorganic acids or ions, and whether they have a significant contribution to the initial formation of molecular clusters, or only to the growth of these clusters.  

 

We have studied clusters of C10-C14 sized accretion products from isoprene and toluene oxidation, as well as clusters of C20 sized accretion products from ⍺-pinene oxidation.  The studied OOMs were obtained   using Gecko-AP, a RO2 + RO2 accretion product generator based on the Gecko-A software.  The main bottleneck for modelling OOM cluster is the conformational sampling of their high-dimensional potential energy surfaces. Thus we we have update previous automated cluster conformational sampling protocols. Initial sampling of cluster configurations was done at semi-empirical level of theory. Minimum free energy configurations were found through successive rounds of filtering and re-optimization at higher DFT levels of theory. As we found that even an extensive sampling of cluster configuration space does not guarantee that the global minimum is found, we introduced constraints to initial sampling which force hydrogen bond formation between molecules. We also used metadynamics simulations to search for additional local minima.  We are currently with neural network potentials which are likely to allow computationally even more effective configurational sampling.

 

The binding free energies of the OOM homodimers are almost uncorrelated with the saturation vapour pressures predicted by existing group-contribution approaches. Binding energy of heterodimers can, however, be estimated from homodimer binding energies with a spread of   ±1-2 kcal/mol, indicating desired tranferability from unimolecular properties to clustering efficiecy. The predicted binding free energies are too high for substantial clustering to occur in typical lower-tropospheric conditions. For validation purposes we performed calculations on dimers of differently sized polyethylene glycol molecules (PEGs), for which the configurational sampling is relatively straightforward, and the saturation vapor pressures are available both from quantum chemistry (via COSMOTherm) and experimentally. Using the PEG molecules, we demonstrate that both the weak binding, and the lack of correlation between binding free energies and saturation vapour pressures, are likely caused by intramolecular hydrogen bonding. This self-bonding is dictated by the molecular flexibility, which is ultimately a unimolecular property, and potentially a cost-effectively descriptor for assessing the clustering ability of OOMs with machine learning based methods.