function of the coordinates of all atoms in the molecular system. All chemically
relevant structures (reactants, intermediates, transition states, and products) are
stationary structures (i.e., zero-gradient) in the PES. Their associated relative
Gibbs energies give the energy landscape of the reaction at given temperature and
pressure, and from their sequence a reaction mechanism is inferred. A recent
example from our production is shown in Fig. 1 [3], and more examples can be
found in the ensuing chapters of this volume, particularly in Chap. 2 [4].
The combination of computational studies with the use of modern experimental
tools has enabled great progress in the understanding of reaction mechanisms
[5, 6]. The continuously increasing computing power combined with the development of efficient and user-friendly software has allowed the full incorporation of
DFT calculations into the toolbox of organometallic chemists’ methods and its
widespread use. However, unlike what happened with other physical methods,
there is no clear awareness of the limitations of the DFT method and its main error
sources. The blind acceptance of computations to interpret reaction mechanisms is
behind recent fierce criticisms of computational studies, in which the authors argue
that computational studies can be wrong and misleading [7, 8].
Before entering into the detailed analysis about what is behind DFT calculations
of energy profiles and what they bring about, some words of caution are in order. The
first issue is the proof (or disproof) of a reaction mechanism. This is a key question to
all chemists interested in how their stochiometric or catalytic reactions work. For
instance, “what constitutes evidence for a catalytic mechanism, as well as how
mechanisms should be evaluated” was a topic of a recent Editorial of ACS Catalysis
[9]. A postulated reaction mechanism is a working hypothesis, whose predictions
must be compared with experiment. Such proposed mechanisms require qualitative
as well as quantitative agreement with experimental observables, and those which
are inconsistent with observations must be discarded. It results that a reaction
mechanism cannot be proven but only disproved [6, 9]. Another point to stress is
that, usually, computational methods do not discover reaction pathways, but instead
evaluate reactions within the scope of existing chemical knowledge [10]. Computational studies of reaction mechanism are often focused on the productive part of the
reaction, but issues like catalyst deactivation or off-cycle reactions generally are not
considered [11]. This implies that only a small part of the PES surface is explored.
Moreover, in general, there are multiple reaction paths connecting the given reactant
and product. The automated reaction path search methods, which have attracted an
increasing attention, overcome these problems and allow the exploration of the full
PES without a prejudgment of the products as well as the reaction paths [10, 12,
13]. A detailed account about this topic is given in Chap. 3 [14]. It must be also
pointed out that computational studies of organometallic reactions have focused
traditionally on the calculation of Gibbs energy profiles, but experiments focus on
reaction rates, which depend also on concentrations. Microkinetic modeling,
consisting in the construction of explicit kinetic reaction networks merging the
rate constants provided by calculations, allows to reproduce the evolution through
time of the reaction species and, therefore, brings data directly comparable to the
What Makes a Good (Computed) Energy Profile?
3
relevant structures (reactants, intermediates, transition states, and products) are
stationary structures (i.e., zero-gradient) in the PES. Their associated relative
Gibbs energies give the energy landscape of the reaction at given temperature and
pressure, and from their sequence a reaction mechanism is inferred. A recent
example from our production is shown in Fig. 1 [3], and more examples can be
found in the ensuing chapters of this volume, particularly in Chap. 2 [4].
The combination of computational studies with the use of modern experimental
tools has enabled great progress in the understanding of reaction mechanisms
[5, 6]. The continuously increasing computing power combined with the development of efficient and user-friendly software has allowed the full incorporation of
DFT calculations into the toolbox of organometallic chemists’ methods and its
widespread use. However, unlike what happened with other physical methods,
there is no clear awareness of the limitations of the DFT method and its main error
sources. The blind acceptance of computations to interpret reaction mechanisms is
behind recent fierce criticisms of computational studies, in which the authors argue
that computational studies can be wrong and misleading [7, 8].
Before entering into the detailed analysis about what is behind DFT calculations
of energy profiles and what they bring about, some words of caution are in order. The
first issue is the proof (or disproof) of a reaction mechanism. This is a key question to
all chemists interested in how their stochiometric or catalytic reactions work. For
instance, “what constitutes evidence for a catalytic mechanism, as well as how
mechanisms should be evaluated” was a topic of a recent Editorial of ACS Catalysis
[9]. A postulated reaction mechanism is a working hypothesis, whose predictions
must be compared with experiment. Such proposed mechanisms require qualitative
as well as quantitative agreement with experimental observables, and those which
are inconsistent with observations must be discarded. It results that a reaction
mechanism cannot be proven but only disproved [6, 9]. Another point to stress is
that, usually, computational methods do not discover reaction pathways, but instead
evaluate reactions within the scope of existing chemical knowledge [10]. Computational studies of reaction mechanism are often focused on the productive part of the
reaction, but issues like catalyst deactivation or off-cycle reactions generally are not
considered [11]. This implies that only a small part of the PES surface is explored.
Moreover, in general, there are multiple reaction paths connecting the given reactant
and product. The automated reaction path search methods, which have attracted an
increasing attention, overcome these problems and allow the exploration of the full
PES without a prejudgment of the products as well as the reaction paths [10, 12,
13]. A detailed account about this topic is given in Chap. 3 [14]. It must be also
pointed out that computational studies of organometallic reactions have focused
traditionally on the calculation of Gibbs energy profiles, but experiments focus on
reaction rates, which depend also on concentrations. Microkinetic modeling,
consisting in the construction of explicit kinetic reaction networks merging the
rate constants provided by calculations, allows to reproduce the evolution through
time of the reaction species and, therefore, brings data directly comparable to the
What Makes a Good (Computed) Energy Profile?
3