of CO2–amine–water systems as a function of composition and
temperature. This VLE system includes the chemical equilibria in
solution which define the solubility of CO2 in the absorbent. Due to
the large concentrations used and associated high ionic strengths it
is expected that the chemical equilibria will behave non-ideally due
to changes in activity of the chemical species. Typically this nonideal
behaviour has been described using complex Pitzer (Bottinger
et al., 2008), UNIQUAC (Aronu et al., 2011) and eNRTL (Zhang
and Chen, 2011) type models based on the assumption that simpler
activity models are inadequate. This assumption has not been
tested. Using these complex activity models requires many tens of
parameters to be known beforehand or determined. These are often
regressed against relatively small experimental datasets resulting
in parameter over-fitting in which many of the parameters are
poorly defined and of little physical relevance. Such over-fitting
can be avoided through the application of techniques such as regularisation,
or cross validation, which give insight into the sensitivity
and robustness of the determined parameters, but this is generally
not done. Furthermore, this over-fitting means such models cannot
be reliably extrapolated which poses a problem for their inclusion
in process models.