306
J. N. Harvey
reasons, we felt emboldened to include the word ‘accurate’ in the title of our paper
[22].
In retrospect, this was unwise, since while some of the mechanistic conclusions
drawn from the study were indeed accurate (such as the observations about enantioselectivity), and some features of our study made it more likely to be accurate than
previous studies, some of the underlying calculated results were very far from being
accurate, as became clear in following years.
First, we decided to go beyond potential energy surfaces and compute free energies, somewhat like has been reported here in Table 1. In fact, hand-waving estimated
of free energies of activation had been included in a footnote in [22] and hinted that the
chosen theoretical protocol overestimated the barrier to reaction. This was emphatically confirmed upon calculating free energies in a new study [23]. The calculated
free energy of the rate-limiting TS using the same DFT functional as in the original
study, B3LYP, was above 50 kcal mol
−1 relative to reactants, and thereby well above
the estimated free energy barrier based on experimentally observed reactivity, which
was of the order of 25 kcal mol
−1 .
The new study simultaneously suggested a resolution of this disagreement for the
case of the non-protic reaction conditions, involving a second equivalent of aldehyde.
A first effect leading to barrier lowering is to change the model of the nucleophile
from trimethylamine to one of the catalysts that is actually used experimentally,
quinuclidine. This is a much better nucleophile, and this change lowers the barrier
by a few kcal mol
−1 . A much more important effect was observed based on local
coupled-cluster ab initio correlated calculations, and on dispersion-corrected B3LYP
(using the same type of Grimme correction already mentioned before in this study): It
appears that B3LYP hugely overestimates the relative potential energy of the key TSs.
This is because the Morita–Baylis–Hillman reaction in this variant involves bringing
four moderately large organic molecules (one tertiary amine, one acrylate ester, and
two aldehydes) together to form a single species. The dispersion effect of roughly
5 kcal mol
−1 shown in Table 1 is for bringing two species together, i.e., forming
one new intermolecular interface. Here, one forms six such interfaces (assuming
each fragment is in contact with each other), or three, assuming a one-dimensional
arrangement—in any case, a much larger effect is expected and indeed found.
The second paper [23] did not go on to consider the reaction under protic conditions, partly due to the large amount of work needed to perform, and critically
assess, the local coupled-cluster calculations for the large system. A second reason
was that the situation in the presence of the alcohol product was considered to be
more complex, so that continuum solvent models were feared to provide a less accurate model of the solvation effects. We did not at all consider reaction in the presence
of a large amount of alcohol, as when the reaction is performed in such solvents.
Reaction in alcohols is not strikingly faster than in polar aprotic solvents including
relatively apolar solvents such as THF or dichloromethane, so we considered that
except for the additional possibility for shuffling shown in Fig. 9, we did not expect
the mechanism to be hugely different in alcohols.
Following both of these papers, a very thorough experimental study [24] was
reported of the Morita–Baylis–Hillman reaction, considering a slightly different vari-
J. N. Harvey
reasons, we felt emboldened to include the word ‘accurate’ in the title of our paper
[22].
In retrospect, this was unwise, since while some of the mechanistic conclusions
drawn from the study were indeed accurate (such as the observations about enantioselectivity), and some features of our study made it more likely to be accurate than
previous studies, some of the underlying calculated results were very far from being
accurate, as became clear in following years.
First, we decided to go beyond potential energy surfaces and compute free energies, somewhat like has been reported here in Table 1. In fact, hand-waving estimated
of free energies of activation had been included in a footnote in [22] and hinted that the
chosen theoretical protocol overestimated the barrier to reaction. This was emphatically confirmed upon calculating free energies in a new study [23]. The calculated
free energy of the rate-limiting TS using the same DFT functional as in the original
study, B3LYP, was above 50 kcal mol
−1 relative to reactants, and thereby well above
the estimated free energy barrier based on experimentally observed reactivity, which
was of the order of 25 kcal mol
−1 .
The new study simultaneously suggested a resolution of this disagreement for the
case of the non-protic reaction conditions, involving a second equivalent of aldehyde.
A first effect leading to barrier lowering is to change the model of the nucleophile
from trimethylamine to one of the catalysts that is actually used experimentally,
quinuclidine. This is a much better nucleophile, and this change lowers the barrier
by a few kcal mol
−1 . A much more important effect was observed based on local
coupled-cluster ab initio correlated calculations, and on dispersion-corrected B3LYP
(using the same type of Grimme correction already mentioned before in this study): It
appears that B3LYP hugely overestimates the relative potential energy of the key TSs.
This is because the Morita–Baylis–Hillman reaction in this variant involves bringing
four moderately large organic molecules (one tertiary amine, one acrylate ester, and
two aldehydes) together to form a single species. The dispersion effect of roughly
5 kcal mol
−1 shown in Table 1 is for bringing two species together, i.e., forming
one new intermolecular interface. Here, one forms six such interfaces (assuming
each fragment is in contact with each other), or three, assuming a one-dimensional
arrangement—in any case, a much larger effect is expected and indeed found.
The second paper [23] did not go on to consider the reaction under protic conditions, partly due to the large amount of work needed to perform, and critically
assess, the local coupled-cluster calculations for the large system. A second reason
was that the situation in the presence of the alcohol product was considered to be
more complex, so that continuum solvent models were feared to provide a less accurate model of the solvation effects. We did not at all consider reaction in the presence
of a large amount of alcohol, as when the reaction is performed in such solvents.
Reaction in alcohols is not strikingly faster than in polar aprotic solvents including
relatively apolar solvents such as THF or dichloromethane, so we considered that
except for the additional possibility for shuffling shown in Fig. 9, we did not expect
the mechanism to be hugely different in alcohols.
Following both of these papers, a very thorough experimental study [24] was
reported of the Morita–Baylis–Hillman reaction, considering a slightly different vari-