Dispersibility of carbon nanotubes in organic solvents: do we really have predictive models?

Predicting the physico-chemical properties of carbon nanotubes (CNTs) is highly demanding owing to tedious experimental efforts involved in their determination. Dispersibility of CNTs in organic solvents is one such property; however, studies involving its quantitative prediction are quite scarce and highly questionable, particularly, when the real external predictivity is desired. This work examines the real external predictivity of the existing models as well as those developed in the present work (using quantum-chemical descriptors) for the dispersibility of single-walled CNTs (SWCNTs). The real external predictivity is assessed on the basis of state-of-the-art external validation parameters obtained by employing an external prediction set which is not exposed to the model used for the prediction. Notably, most of the present and existing models pass through the internal validation, but unfortunately, barring some exception of poly-parameter models, most of the models fail when it comes to the external validation. Their failure was attributed to the descriptors employed which are in fact based on the gas-phase single molecular structure of organic solvents as well as based on the implicit solvent methods. A future approach towards the predictive models for the dispersibility of CNTs is suggested based on the explicit solvent methods.