308
J. N. Harvey
acknowledged in the computational community. Post hoc rationalizations of experimental observations are much more common. The suggestion that basic organic
chemistry knowledge is already enough to identify the mechanism has some merit,
and indeed I already noted near Fig. 8 that the broad features of the mechanism are
obvious from basic knowledge of organic chemistry, with the key discussion concerning the way in which the proton shuffling occurs. Also, it has been noted that
the calculated energies in our first study, derived from B3LYP, were in error by more
than 20 kcal mol
−1 for some species, which is a huge error. For these reasons, the
criticisms of the previous computational studies were fully warranted.
However, it should be noted that the mechanism as shown in Fig. 10 cannot in
fact be the only mechanism for the reaction, since it explicitly requires the presence
of a proton-donating alcohol solvent, while the reaction can be performed in nonprotic solvents, or using a neat mixture of the (non-protic) reactants. Hence, another
mechanism—the two-aldehyde mechanism mentioned above and considered in both
of our computational studies [22, 23]—must be favored under some conditions (it is
not completely clear under which conditions which of the mechanisms is favored).
It should also be noted that our own work had in fact never set out to predict the
mechanism in pure alcohol solvent (the alcohol that we did consider was the product, although it was modeled as methanol). Finally, we note that the conclusions
surrounding enantioselectivity are much less easy to derive simply using paper and
pencil, and knowledge of organic reaction mechanisms.
While retrodiction is much less satisfying than prediction, the publication of [24]
did spur us on (together with another colleague, Ragahavan Sunoj, who had also
previously published on the Morita–Baylis–Hillman reaction) to revisit the reaction
and attempt to examine computationally the mechanism for the precise variant of the
reaction studied in the new study [24], i.e., with DABCO catalyst, methyl acrylate,
and p-nitrobenzaldehyde, in methanol. Using local coupled-cluster methods, the free
energy surface emerging from this new study [16] is shown in Fig. 11. As can be seen
our calculated free energies are in fair agreement with experiment, with a maximum
error of just under 5 kcal mol
−1 , in line with the expected accuracy of the methods
used.
Our new study [16] also aimed to examine the origin of the errors in earlier studies.
Some of these are obvious: The poor description of dispersion in B3LYP has already
been mentioned several times in this chapter and leads to hugely inaccurate energies
for this reaction. It has also already been mentioned that some of the ‘error’ in some
of the studies comes from the fact that the mechanism of the Morita–Baylis–Hillman reaction is different under different reaction conditions. We also showed that
changes in the solvent model (various continuum models or a free energy perturbation
method with molecular mechanics), or in the treatment of the entropy arising from
soft vibrational modes, or changing from one more accurate dispersion-corrected
DFT functional to another or to coupled-cluster theory, are enough to account for
changes in relative free energies of several kcal mol
−1 . The Morita–Baylis–Hillman
reaction involves very large changes in the solvation free energy, so it may represent a
more challenging target for accurate computational studies than some other reactions
involving less change in polarity along the reaction path.
J. N. Harvey
acknowledged in the computational community. Post hoc rationalizations of experimental observations are much more common. The suggestion that basic organic
chemistry knowledge is already enough to identify the mechanism has some merit,
and indeed I already noted near Fig. 8 that the broad features of the mechanism are
obvious from basic knowledge of organic chemistry, with the key discussion concerning the way in which the proton shuffling occurs. Also, it has been noted that
the calculated energies in our first study, derived from B3LYP, were in error by more
than 20 kcal mol
−1 for some species, which is a huge error. For these reasons, the
criticisms of the previous computational studies were fully warranted.
However, it should be noted that the mechanism as shown in Fig. 10 cannot in
fact be the only mechanism for the reaction, since it explicitly requires the presence
of a proton-donating alcohol solvent, while the reaction can be performed in nonprotic solvents, or using a neat mixture of the (non-protic) reactants. Hence, another
mechanism—the two-aldehyde mechanism mentioned above and considered in both
of our computational studies [22, 23]—must be favored under some conditions (it is
not completely clear under which conditions which of the mechanisms is favored).
It should also be noted that our own work had in fact never set out to predict the
mechanism in pure alcohol solvent (the alcohol that we did consider was the product, although it was modeled as methanol). Finally, we note that the conclusions
surrounding enantioselectivity are much less easy to derive simply using paper and
pencil, and knowledge of organic reaction mechanisms.
While retrodiction is much less satisfying than prediction, the publication of [24]
did spur us on (together with another colleague, Ragahavan Sunoj, who had also
previously published on the Morita–Baylis–Hillman reaction) to revisit the reaction
and attempt to examine computationally the mechanism for the precise variant of the
reaction studied in the new study [24], i.e., with DABCO catalyst, methyl acrylate,
and p-nitrobenzaldehyde, in methanol. Using local coupled-cluster methods, the free
energy surface emerging from this new study [16] is shown in Fig. 11. As can be seen
our calculated free energies are in fair agreement with experiment, with a maximum
error of just under 5 kcal mol
−1 , in line with the expected accuracy of the methods
used.
Our new study [16] also aimed to examine the origin of the errors in earlier studies.
Some of these are obvious: The poor description of dispersion in B3LYP has already
been mentioned several times in this chapter and leads to hugely inaccurate energies
for this reaction. It has also already been mentioned that some of the ‘error’ in some
of the studies comes from the fact that the mechanism of the Morita–Baylis–Hillman reaction is different under different reaction conditions. We also showed that
changes in the solvent model (various continuum models or a free energy perturbation
method with molecular mechanics), or in the treatment of the entropy arising from
soft vibrational modes, or changing from one more accurate dispersion-corrected
DFT functional to another or to coupled-cluster theory, are enough to account for
changes in relative free energies of several kcal mol
−1 . The Morita–Baylis–Hillman
reaction involves very large changes in the solvation free energy, so it may represent a
more challenging target for accurate computational studies than some other reactions
involving less change in polarity along the reaction path.