310
J. N. Harvey
be exaggerated. This suggestion was motivated by several studies [26] arguing that
experimental entropies of solvation are quite negative, so that solutes have a lower
entropy than suggested by the Sackur–Tetrode equation. These negative entropies of
solvation are indeed frequently observed experimentally, but their existence does not
mean that one should not use a version of the Sackur–Tetrode equation to compute
entropies hence free energies in solution. Indeed, the negative entropy of solvation
should be accounted for by the use of the continuum solvent model, which is parameterized to reproduce the experimental free energy of solvation, including both its
enthalpic and entropic factors. Seeking to make a separate adjustment of the solventphase entropy would lead to a double-counting of solvation entropies. In our new
study [16], the ‘full’ Sackur–Tetrode entropy is included, and the good agreement
with experiment vindicates the present argument. A recent study [27] has compared
various ways to compute solvent-phase entropies and also argues against making
substantial ad hoc adjustments to the Sackur–Tetrode approach.
4 Conclusions
The above examples show that computational chemistry can make important contributions to understanding reaction mechanisms in organic and organometallic chemistry. However, the examples described show that obtaining accurate results is much
more challenging than had been appreciated when I started to carry out work in this
area. The chapter outlines many of the pitfalls that we fell into along the way—or
that we avoided due to good luck, and the vigilance of referees or experimental
collaborators, rather than anything else.
What are some of the key lessons learned? First is that electronic structure theory
remains challenging. It is to be hoped that there will be no future discovery of
unexpected but very large errors in a commonly used method, such as the discovery of
the magnitude of the error made by using non-dispersion-adjusted functionals such as
B3LYP that occurred over the period covered by the research described in this chapter.
In retrospect, the discovery of the importance and magnitude of this error for organic
and organometallic mechanistic chemistry must be considered to be enormously
chastening for the field. It occurred due to the fact that the chemical space where
functionals such as B3LYP had been widely benchmarked—largely, a space occupied
only by small molecules—ceased to overlap with the regions of chemical space where
it was being used, as more powerful computers enabled application to larger and
larger species. This observation invites caution for the future, as computers continue
to expand in capacity, so more and more systems of more and more varied properties
are studied. My view is that the future will contain more chastening experiences
where a given method is discovered to be much less accurate than expected for a
problem to which it has applied and where results have been published.
A second observation and recommendation relates to the fact that while calculating
potential energy surfaces and free energies is definitely challenging, deducing the
kinetic and thermodynamic conclusions of these calculations can also be surprisingly
J. N. Harvey
be exaggerated. This suggestion was motivated by several studies [26] arguing that
experimental entropies of solvation are quite negative, so that solutes have a lower
entropy than suggested by the Sackur–Tetrode equation. These negative entropies of
solvation are indeed frequently observed experimentally, but their existence does not
mean that one should not use a version of the Sackur–Tetrode equation to compute
entropies hence free energies in solution. Indeed, the negative entropy of solvation
should be accounted for by the use of the continuum solvent model, which is parameterized to reproduce the experimental free energy of solvation, including both its
enthalpic and entropic factors. Seeking to make a separate adjustment of the solventphase entropy would lead to a double-counting of solvation entropies. In our new
study [16], the ‘full’ Sackur–Tetrode entropy is included, and the good agreement
with experiment vindicates the present argument. A recent study [27] has compared
various ways to compute solvent-phase entropies and also argues against making
substantial ad hoc adjustments to the Sackur–Tetrode approach.
4 Conclusions
The above examples show that computational chemistry can make important contributions to understanding reaction mechanisms in organic and organometallic chemistry. However, the examples described show that obtaining accurate results is much
more challenging than had been appreciated when I started to carry out work in this
area. The chapter outlines many of the pitfalls that we fell into along the way—or
that we avoided due to good luck, and the vigilance of referees or experimental
collaborators, rather than anything else.
What are some of the key lessons learned? First is that electronic structure theory
remains challenging. It is to be hoped that there will be no future discovery of
unexpected but very large errors in a commonly used method, such as the discovery of
the magnitude of the error made by using non-dispersion-adjusted functionals such as
B3LYP that occurred over the period covered by the research described in this chapter.
In retrospect, the discovery of the importance and magnitude of this error for organic
and organometallic mechanistic chemistry must be considered to be enormously
chastening for the field. It occurred due to the fact that the chemical space where
functionals such as B3LYP had been widely benchmarked—largely, a space occupied
only by small molecules—ceased to overlap with the regions of chemical space where
it was being used, as more powerful computers enabled application to larger and
larger species. This observation invites caution for the future, as computers continue
to expand in capacity, so more and more systems of more and more varied properties
are studied. My view is that the future will contain more chastening experiences
where a given method is discovered to be much less accurate than expected for a
problem to which it has applied and where results have been published.
A second observation and recommendation relates to the fact that while calculating
potential energy surfaces and free energies is definitely challenging, deducing the
kinetic and thermodynamic conclusions of these calculations can also be surprisingly