Computer-aided rational design of Fe(iii)-catalysts for the selective formation of cyclic carbonates from CO2 and internal epoxides

Article abstract of DOI:10.1039/c7cy01435j

The catalytic mechanism of the cyclic carbonate formation reaction between CO₂ and internal epoxides promoted by Fe-salen and the Kleij catalyst was examined in detail to better understand how the catalytic efficiency can be increased. Specifically, we aimed to make the catalyst more chemoselective towards forming cyclic carbonates and preventing the competing side reaction leading to polycarbonates via ring-opening polymerization. A few rational design principles were derived and first tested using computer models based on density functional theory. The most promising candidate that was identified in the computer model was then prepared and found to display significantly enhanced reactivity towards forming the cyclic carbonates, supporting the validity of the mechanistic insights deduced from the computer simulations. We propose that a cyclic carbonate is formed most efficiently via an inner-sphere mechanism where both the CO₂ and epoxide substrates utilize the metal center for the key bond forming events. In contrast, the ring-opening polymerization uses an outer-sphere mechanism, where a carbonate attacks and ring-opens the epoxide bound to the metal without engaging the metal directly. These mechanistic differences are exploited to implement a chemoselective catalyst by enhancing the rate of the cyclic carbonate formation reaction while leaving the polymerization pathway largely unaffected.

Full text of DOI:10.1039/c7cy01435j

View Article Online
View Journal
Catalysis
Science &
Technology
Accepted Manuscript
This article can be cited before page numbers have been issued, to do this please use: M. Baik, I. Sinha, S.
Lee, C. Bae, S. Tussupbayev, Y. Lee, M. Seo, J. Kim, Y. Lee and H. Kim, Catal. Sci. Technol., 2017, DOI:
10.1039/C7CY01435J.
This is an Accepted Manuscript, which has been through the
Volume 6 Number 1 7 January 2016 Pages 1–308
Royal Society of Chemistry peer review process and has been
accepted for publication.
Catalysis
Science &
Technology
Accepted Manuscripts are published online shortly after
www.rsc.org/catalysis
acceptance, before technical editing, formatting and proof reading.
Using this free service, authors can make their results available
to the community, in citable form, before we publish the edited
article. We will replace this Accepted Manuscript with the edited
and formatted Advance Article as soon as it is available.
You can find more information about Accepted Manuscripts in the
author guidelines.
Please note that technical editing may introduce minor changes
to the text and/or graphics, which may alter content. The journal’s
ISSN 2044-4753
standard Terms & Conditions and the ethical guidelines, outlined
in our author and reviewer resource centre, still apply. In no
event shall the Royal Society of Chemistry be held responsible
PAPER
Qingzhu Zhang et al.
Catalytic mechanism of C–F bond cleavage: insights from QM/MM
analysis of fluoroacetate dehalogenase
for any errors or omissions in this Accepted Manuscript or any
consequences arising from the use of any information it contains.
rsc.li/catalysis

Page 1 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
ComputerꢀAided Rational Design of Fe(III)ꢀCatalysts for the Selective Formation of
Cyclic Carbonates from CO2 and Internal Epoxides
Indranil Sinha,†,‡,ǁ ,Yuseop Lee,‡,ǁ Choongman Bae,‡ Samat Tussupbayev,†,‡ Yujin Lee,†,‡ MinꢀSeob Seo,‡
Jin Kim,‡ MuꢀHyun Baik,†,‡,* Yunho Lee†,‡,* and Hyunwoo Kim‡,*
†Center for Catalytic Hydrocarbon Functionalizations, Institute for Basic Science (IBS), Daejeon 34141,
South Korea.
‡Department of Chemistry, Korea Advanced Institute of Science and Technology (KAIST), Daejeon
34141, South Korea.
KEYWORDS: catalyst design, Fe catalyst, cyclic carbonate, carbon dioxide
Abstract
The catalytic mechanism of the cyclic carbonate forming reaction between CO2 and internal
epoxides promoted by Feꢀsalen and the Kleij catalyst was examined in detail to better understand how
the catalytic efficiency can be increased. Specifically, we aimed to make the catalyst more
chemoselective towards forming cyclic carbonates and preventing the competing side reaction leading to
polycarbonates via ringꢀopening polymerization. A few rational design principles were derived and first
tested using computer models based on density functional theory. The most promising candidate that
was identified in the computer model was then prepared and found to display significantly enhanced
reactivity towards forming the cyclic carbonates, supporting the validity of the mechanistic insights
deduced from the computer simulations. We propose that the cyclic carbonate is formed most efficiently
in an innersphere mechanism where both the CO2 and epoxide substrates utilize the metalꢀcenter for the
key bond forming events. In contrast, the ringꢀopening polymerization uses an outersphere mechanism,
where a carbonate attacks and ringꢀopens the epoxide bound to the metal without engaging the metal
directly. These mechanistic differences are exploited to implement a chemoselective catalyst by
enhancing the rate of the cyclic carbonate forming reaction, while leaving the polymerization pathway
largely unaffected.
1

Catalysis Science & Technology
Page 2 of 22
View Article Online
DOI: 10.1039/C7CY01435J
Introduction
Concerns about depleting petroleum reserves and the environmental problems that result from
hydrocarbon combustion continue to drive research efforts towards renewable and clean alternatives to
fossil fuels. An equally serious problem that must be solved as we progress towards a carbonꢀneutral
economy is that we must secure alternative sources for feedstock chemicals that are currently derived
from petroleum. One attractive solution is to use atmospheric CO2 and develop chemical reactions that
will capture and transform it into value added feedstock chemicals.1 Specifically, coupling of CO2 and
epoxides to form polycarbonates or cyclic carbonates is a promising reaction for industrial utility
because some of these products are already valuable raw materials for functional polymers and they
could serve as monomers for polycarbonate synthesis, electrolytes, and polar solvents.1ꢀ3 Many metalꢀ
catalysts2 and organocatalysts3 have been reported to carry out these reactions. For example, salenꢀ
complexes (salen = N,N'ꢀbis(salicylidiene)ethylenediamine) of Al, Cr, Mn, Co, Ti and Zn are well
known for catalyzing the formation of polycarbonates or cyclic carbonates, as illustrated in Scheme 1.2,4
[Cr(salen)]5 and [Co(salen)]6 complexes were used for polycarbonate synthesis, whereas binuclear
[Al(salen)]2O7 was used for cyclic carbonate synthesis. One of the unsolved challenges lies in
controlling the chemoselectivity of these catalysts. As all reactions coupling CO2 and epoxide are
necessarily carried out in the presence of the two said substrates, there is always a competition between
the cyclic carbonate and the polycarbonate forming reaction pathways.
Scheme 1 (i) Coupling of CO2 and epoxides to produce cyclic carbonates or polycarbonates catalyzed
by metal complexes based on (ii) salen (1a), (iii) amino tris(phenolate) (1b), and (d) a new N1–O3 ligand
(1c).
Recently, Kleij et al. reported that cyclic carbonates can be formed with high selectivity using
Al9 and Fe10 complexes based on aminoꢀtris(phenolate) ligands, shown as 1b in Scheme 1. These
catalysts were reported to be highly efficient and selective for the formation of cyclic carbonates from
2

Page 3 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
terminal epoxides, but our experiments indicated that the selectivity was significantly reduced when
internal epoxides were used. Unfortunately, there is little understanding of how such selective behavior
is achieved and, as a result, there is no rational strategy for improving the chemoselectivity. The
stereochemical orientation of ligand in 1b was proposed to be important, but it is not clear which
features of the ligands are ultimately responsible in this class of catalysts. In principle, developing
catalysts that are chemoselective for either cyclic carbonates or polycarbonates is hard, because the
fundamental reaction steps that lead to the two possible products are similar, if not identical. Quantum
chemical reaction modeling studies provided some inspiration, but have not resulted in an exploitable
design strategy.11 We wished to derive a deeper mechanistic understanding of the catalytic reactions
leading to the cyclic carbonate and the polycarbonate products, such that we can identify an exploitable
strategy for obtaining a highly chemoselective and efficient new catalyst.
Results and Discussions
The ultimate goal of this study is to rationally design a highly chemoselective catalyst that can
produce cyclic carbonate products in high yields while avoiding polymeric side products based on a
deep understanding of the mechanism. In Kleij’s inspirational works,10 Fe was shown to be a competent
metal for the cyclic carbonate formation with terminal epoxides, but interestingly, only very modest
chemoselectivity was observed with internal epoxides, as will be discussed below. Although some
notable reactivity was also observed with Al, we chose to focus on iron, as we speculated that the
electronic structure dominated by dꢀorbitals should be more versatile in principle and the chemical
reactivity of the iron complex should be more responsive to structural changes of the ligand than the Alꢀ
system.
CO2 Activation by Fe-Catalysts. To establish a prototype mechanism, the formation of a cyclic
carbonate using the “firstꢀgeneration catalyst” Fe(salen)Cl was examined. It was previously reported to
be a poor catalyst for the copolymerization of carbon dioxide and propylene oxide.12 The general
catalytic cycle envisioned for all iron complexes considered in this study is illustrated in Scheme 2. The
reactant complex 2 carrying an epoxide substrate may be attacked first by a bromide anion in a
nucleophilic SN2 reaction, which gives the ringꢀopened intermediate 3. Subsequent insertion of CO2 into
the Fe–O bond may afford the carbonate intermediate 4 traversing the transition state 3-TS, where we
speculate that the metal site may stabilize the negative charge polarization on the terminal oxygen, as
illustrated in Scheme 2. Extrusion of the bromide accompanied by C–O bond formation may lead to the
cyclic carbonate 5. To test the catalytic activity of Fe(salen)Cl towards the cyclic carbonate formation,
3

Catalysis Science & Technology
Page 4 of 22
View Article Online
DOI: 10.1039/C7CY01435J
we prepared and exposed it to standard reaction conditions. In good agreement with the previous
report,12 we found that Fe(salen)Cl is not at all an effective catalyst demonstrating a low conversion rate,
that was never higher than ~30%.
Scheme 2 Proposed mechanistic cycle for Fe(III)catalyzed coupling of cyclic epoxides and CO2, that
has been computationally investigated.
The calculated reaction energy profile is presented in Figure 1 and shows that the epoxide ring
opening accompanied by the nucleophilic attack by the bromide ion is viable with a reaction barrier of
only 16.5 kcal/mol and the ringꢀopened intermediate 3a is predicted to readily form having a relative
free energy of only 11.1 kcal/mol. The ironꢀassisted carbon dioxide insertion step, however, is
associated with the transition state 3a-TS that was located at 45.6 kcal/mol. Such a prohibitively high
barrier is unexpected, although it would be in qualitative agreement with the experimentally observed
poor catalytic performance. A more detailed inspection reveals the reason for such a high barrier. As
shown in Figure 2, the computed structure of 3a-TS necessarily consists of a 7ꢀcoordinate ironꢀcomplex.
Throughout the reaction the iron center remains in the high spin manifold and it is of course particularly
challenging to access such a high coordination number with a highꢀspin metal center at a relatively low
valent Fe(III) state, where the highest occupied frontier orbitals are metal–ligand antibonding in
character. All plausible alternative spin states were found to be too high in energy to be chemically
relevant. For example, the lowꢀspin doublet analogue of the reactant complex 2a is 31.2 kcal/mol higher
in energy than the Fe(III)ꢀd5 sextet ground state.
4

Page 5 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
∆G(sol)
[kcal/mol]
Br
O
O
C
O
Fe
O
O
Br
3a-TS
(45.57)
•
C
O
50
40
Br
O
O
Fe
Br
4a-TS
(33.97)
•
C
O
O
Fe
Fe
3a-TS'
(27.60)
•
30
20
10
0
2a-TS
(16.54)
*
O
O
•
C O
•
*
4a
O
(16.96)
•
3a
Fe
(11.09)
Fe
O
•
5a
•
•
O
Br
2a
(2.15)
O
(0.00)
O
Br
Fe
Fe
Figure 1. Energy profile for the formation of the cyclic carbonate product using Fe(salen)Cl, 1a, as the
catalyst.
Figure 2. DFT optimized geometries of the carbon dioxide insertion transition states 3a-TS and 3a-TS'.
Hydrogen atoms were omitted for clarity.
After some exploration, we found that there is a lower energy transition state for the carbon
dioxide insertion that, surprisingly, does not engage the metal center. Instead, the carbon dioxide
substrate may attack the oxo moiety directly via the transition state 3a-TS', as shown in Figure 2,
forming the carbonate while severing the Fe–O bond. This dissociative mechanism requires a rebound
action of the newly formed carbonate to afford intermediate 4a and it is clear that such dissociative step
is potentially highly inefficient, as the carbonate may fail to properly reꢀbind to the iron center and
simply diffuse away from the catalytic site. We propose that it is this inefficiency that gives rise to the
low catalytic activity.
5

Catalysis Science & Technology
Page 6 of 22
View Article Online
DOI: 10.1039/C7CY01435J
Therefore, the prototype mechanism discussed above provides an obvious reason for the
improved catalytic activity displayed by Kleij’s catalyst 1b, shown in Scheme 1. As the aminoꢀ
tris(phenolate) ligand generates an iron center in pseudoꢀoctahedral coordination geometry with two
empty ligand binding sites, it is easy to envision how the axial position will bind the epoxide substrate
and the free equatorial binding site may be used to assist the CO2 insertion in an associative mechanism.
This reaction pathway is expected to be much more efficient by resolving the problems associated with
the rebound step that was pointed out above. Interestingly, our calculations show that such an
associative path leads to a barrier of 27.7 kcal/mol traversing the insertion transition state 3b-TS, as
illustrated in Figure 3. This barrier is identical to what was found for the Feꢀsalen complex. Thus, Kleij’s
catalyst provides no energetic advantage for this insertion step. The intermediate 4b is 15.0 kcal/mol
uphill energetically from the initial reactant 2b. The cyclization of the carbonate and elimination of
bromide to give the product complex 5b is associated with a barrier of 32.3 kcal/mol. This barrier
estimate is slightly too high given that the Kleij catalyst is able to produce cyclic carbonate at 35 °C with
a yield of 52% in 5 hours. Significant efforts were made to identify a lower energy transition state,
including various different orientations of the CO2 and the alkoxide ring, but we were unable to find any
transition state that was lower in energy. Thus, the DFTꢀestimated transition states are a bit too high, but
are still in a reasonable range and our calculations predict that this nucleophilic cyclization step is rate
determining. Although 4b-TS is ~1.7 kcal/mol lower in energy than the Feꢀsalen system 4a-TS, this
energy difference is too small to confidently state that this is the reason for the increased reactivity of
Kleij’s catalyst.
Figure 3. Energy profile of the cyclic carbonate formation using Kleij’s catalyst 1b.
6

Page 7 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
In conclusion, our comparative mechanistic study of the Feꢀsalen and Kleij’s catalysts teaches
the following lessons. First, the coordinatively unsaturated geometry of the iron center allows for adding
an epoxide at the axial position and providing an additional binding site for the metalꢀassisted insertion
of carbon dioxide. We propose that this innersphere mechanism is critically important for increasing the
activity of the metal catalyst by enhancing the robustness of the catalytic reaction. It ensures that the two
substrate molecules are anchored tightly to the metal center throughout the reaction. This feature must
be maintained, therefore, when designing a new catalyst. Another lesson is that a potential design
strategy for preferentially making the cyclic carbonate and reducing the polycarbonate formation
significantly lies in decreasing the binding energy of the second epoxide in the equatorial position.
Several strategies were considered, but two modifications were found to be particularly promising. First,
converting the amine at the axial binding site to an imine functionality to afford a salicylidene group will
increase the electron density at the metal center, rendering it a weaker Lewis acid and thus making it less
attractive for the incoming Lewis base, the epoxide. Second, shifting one of the phenolates away from
the amineꢀnitrogen anchor, as shown in Figure 4, restricts the ability of the tris(phenolate) chelating
ligand to change the bonding angles between the equatorial oxygen ligands when the second epoxide
substrate approaches the metal center. This structural modification should also increase the
chemoselectivity towards the cyclic carbonates by discouraging the binding of the second substrate.
Figure 4. DFT optimized geometries of 1b and 1c. Hydrogen atoms were omitted for clarity.
7

Catalysis Science & Technology
Page 8 of 22
View Article Online
DOI: 10.1039/C7CY01435J
G(sol)
[kcal/mol]
O
Br
C
O
Br
O
C
O
Fe
O
O
Fe
4c-TS
(28.20)
•
Br
30
3c-TS
(25.53)
O
•
Fe
20
10
0
O
O
2c-TS
(9.78)
•
C
O
O
•
4c
•
(9.83)
Fe
Fe
3c
•
•
(4.66)
5c
(0.78)
2c
O
(0.00)
O
Br
O
O
Br
Fe
Fe
Figure 5. Energy profile of the cyclic carbonate formation catalyzed by 1c.
Figure 5 shows the computed reaction energy profile for the new catalyst. The initial epoxide
ring opening reaction is greatly enhanced by the presence of the much stronger donating sp2ꢀnitrogen of
the imine functionality in the axial position compared to the sp3ꢀnitrogen of the amine ligands, which
weakens the O–C bond of the epoxide more effectively and affords a very low free energy of activation
of only 9.8 kcal/mol associated with 2c-TS. In the other two catalysts, this step was associated with a
barrier of ~15 kcal/mol. This transꢀeffect can be further highlighted by further modulating the electronic
demand: We calculated the transition state of a putative catalyst by adding to 2b an electronꢀ
withdrawing –NO2 group at the metaꢀ and an electronꢀdonating –OCH3 group at the paraꢀposition,
respectively. The electronic energies of the transition states were 11.86 and 16.02 kcal/mol, which are
predictably lower and higher than the electronic energy of 14.02 kcal/mol of 2c-TS. While this is an
initially unanticipated feature, it is a welcome and logical consequence of having turned the amino into
an imino ligand. The carbon dioxide insertion is also notably more favorable with a barrier of 25.5
kcal/mol, which is ~2 kcal/mol lower in energy than what was seen with Kleij’s catalyst. Although this
energy difference is smaller than what we had anticipated, the conceptual idea and numerical
confirmation are encouraging. Perhaps most importantly, the cyclization step, which was identified to be
rate limiting, is predicted to proceed via 4c-TS at 28.2 kcal/mol, which is nearly 6 kcal/mol lower in
energy than the corresponding transition state 4b-TS in the known catalyst system. Thus, the proposed
minor modifications to the chelating ligand are predicted to lead to substantial improvement of the
catalytic reaction that affords the cyclic carbonate product. It is important to note here that whereas we
had applied a design logic from our initial mechanistic analysis, the details and extent of the energy
changes cannot be predicted without the specific calculations. These observations are therefore a
8

Page 9 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
convincing example of combining an iterative analysis and exploration of computed reaction energy
profiles to identify a promising catalyst optimization strategy.
Polymerization Mechanism. Instead of undergoing the cyclization described above, 4c can be
engaged by another equivalent of epoxide to initiate the ringꢀopening polymerization reaction, as
illustrated in Figure 6. The ringꢀopening can proceed in an intraꢀ or intermolecular fashion giving rise to
two similar, yet fundamentally different pathways. The intermolecular path involves 2ꢀbromocyclohexyl
carbonate liberation to generate the 15ꢀelectron complex 2c followed by the ringꢀopening step
processing via the transition state 2'c-TS, where the carbonate attacks the coordinated epoxide. In the
alternative path the carbonate attacks the epoxide intramolecularly traversing the transition state 4'c-TS.
Our calculations predict that the intermolecular pathway is kinetically favored over the intramolecular
mechanism. Namely, the ringꢀopening transition state 2'c-TS is 16 kcal/mol lower in energy than 4'c-TS.
This decisive difference in energies can be understood based on their optimized geometries depicted in
Figure 7. Comparing the two transition states it is easy to see that binding the carbonate at the axial
position prior to the epoxide ringꢀopening, restricts the possible SN2 attack geometries. In 4'c-TS, the
O–C–O vector of the substitution reaction is a part of a highly distorted sixꢀmembered ring. Meanwhile,
in the energetically much more favorable 2'c-TS the free carbonate can adopt any geometry and it forms
a relaxed and classical linear geometry along the O–C–O vector that is expected for a SN2 reaction. As
such, our initial guiding principle mentioned above that sought to discourage the binding of the second
epoxide and discriminate the polymerization reaction in this way is not an effective strategy. This
finding highlights the importance of precisely knowing the mechanism before a viable design strategy
can be devised. In this case, the intermolecular nature of the polymerization mechanism provides only a
few opportunities for control and, in principle, a very different strategy would be needed.
The computed reaction energy profiles suggest that the polycarbonate producing reaction is
generally less favorable in comparison to the cyclic carbonate formation with a rate limiting barrier of
~36 kcal/mol. As can be seen from Figure S15 the intermolecular pathway to polycarbonate formation
for Kleij’s catalyst is very similar to what is presented in Figure 6 for 1c. The rate limiting barrier for
Kleij’s catalyst is ~34 kcal/mol. This observation provides a plausible explanation for the fact that the
polymerization reaction is competitive for the Kleij’s catalyst, as the cyclic carbonate formation was
found to be associated with a barrier of ~32 kcal/mol. Using the new catalyst complex 1c, we were able
to lower the cyclic carbonate forming barrier to ~28 kcal/mol, while leaving the polymerization barrier
unaffected at ~36 kcal/mol, as described above. Thus, we may expect a significantly improved
chemoselectivity towards cyclic carbonate formation. In the case of Feꢀsalen, the lack of a second
epoxide binding site in the equatorial position makes the polymerization pathway highly inefficient, as
9

Catalysis Science & Technology
Page 10 of 22
View Article Online
DOI: 10.1039/C7CY01435J
the growing polycarbonate chain cannot remain bound while a new epoxide monomer is brought in
close proximity to it by the metal using the second available coordination site, as shown in 4'c. Thus, we
expect that Feꢀsalen should show very inefficient polymerization activity and, as a result, show high
selectivity for the cyclic carbonate formation reaction pathway.
Br
∆
G(sol)
[kcal/mol]
O
60
O
O
O
O
O
O
Br
4'c-TS
(51.49)
Fe
O
50
40
30
20
10
Fe
2'c-TS
(35.65)
6'c
(30.96)
Intramolecular
*
*
Intermolecular
*
O
O
Br
2'c
O
(25.00)
2c
(20.83)
*
Fe
4'c
(15.78)
O
Fe
4c
(9.83)
6c
(10.57)
O
O
O
O
O
O
Br
Br
Fe
O
O
Fe
Figure 6. Free energy profile of the initial phase of the ringꢀopening polymerization reaction catalyzed
by 1c.
Figure 7. DFT optimized geometries of the ringꢀopening transition state 2'-TS and 4'-TS. Hydrogen
atoms were omitted for clarity.
Encouraged by these computer model studies, we sought to prepare two versions of the new
ligands c and d. They were readily prepared by a scalable threeꢀstep procedure depicted in Scheme 3.
10

Page 11 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
2,2'ꢀdihydroxybenzophenone (7) reacted with ammonia at ambient temperature to form the
corresponding imine, 2,2'ꢀ(iminomethylene)diphenol (8) in 90% yields. In general, imines formed from
ammonia are too unstable to be isolated, probably due to the internal hydrogen bonds between the imine
and two phenol groups stabilizing 8. Subsequent reduction with NaBH4 afforded bis(2ꢀ
hydroxyphenyl)methanaminium chloride (9), isolated as a salt. The final step is the Schiff base
formation between 9 and required salicylaldehyde derivatives, which gave 10c and 10d in 84% and 87%
yields, respectively. This straightforward threeꢀstep procedure allows not only a facile tuning of the
ligand structure but also multigram scale production (~50 g) without any tedious purification techniques.
Single crystals were grown by slow evaporation of pentane into a saturated THF solution of 10c or 10d
at room temperature and the molecular structures were determined by Xꢀray diffraction analysis, which
was found to be the zwitterionic form shown in Figure 8. The phenolic proton of a salicylidene group is
transferred to the imine nitrogen atom resulting in an intramolecular N–H···O hydrogen bond. The
hydrogen bonds found in both 10c (N–H···O: 0.87(2) Å; 1.89(2) Å; 139(1)º) and 10d (N–H···O: 0.99(3)
Å; 1.83(3) Å; 134(3)º) are in good agreement with that of the reported phenolateꢀiminium zwitterionic
form.13 This kind of resonanceꢀassisted hydrogen bond13 is strong and is commonly used to prepare a
variety of chiral vicinal diamines.14
Scheme 3 Synthesis of N1–O3 ligands 10c and 10d and their Fe(III) complexes 1c and 1d.
11

Catalysis Science & Technology
Page 12 of 22
View Article Online
DOI: 10.1039/C7CY01435J
Figure 8. Displacement ellipsoid (50%) representations of (a) 10c and (b) 10d. All C–H hydrogen atoms
and a coꢀcrystallized solvent molecule are omitted for clarity. Dotted lines denote intramolecular
hydrogen bonds. See Supporting Information for details.
The desired Fe(III) catalysts were prepared using the new ligands 10c and 10d. All Fe(III)
complexes were extensively characterized and their molecular structures were determined by Xꢀray
crystallography. The ligands were reacted with FeCl3 in THF to liberate HCl and form neutral
complexes 1c and 1d in good yields. Detailed experimental procedures are given in the Supporting
Information. The Xꢀray diffraction analysis shows interestingly, that both Fe(III) complexes are dimeric
in solid state, having the molecular formula [1c(THF)]2 and [1d]2, in which two metal centers are
bridged by two phenolate oxygen donors from two different ligands, as shown in Figure 9. Few
examples of structurally characterized dimeric Fe(III) complexes supported by a N1–O3 chelating
ligands were reported previously.9,11,15 Both metal centers in each dimeric forms [1c(THF)]2 and [1d]2
are crystallographically equivalent. Two bridging oxygen atoms between the Fe(III) centers originate
from the two phenolate moieties attached to the triꢀsubstituted carbon arm and not from the salicylidene
group. M–Obridging bond distances are fairly similar for these complexes: 2.077(1) vs. 2.020(1) Å for
[1c(THF)]2 and 1.977(2) vs. 2.029(2) Å for [1d]2, suggesting that both ligands adequately accommodate
a symmetrical metal coordination environment (Figure 9 and Table 1). The structure of [1c(THF)]2
shows an octahedral geometry for each iron center, while the analogous [1d]2 structure reveals an
intermediate geometry between square pyramid (SP) and trigonal bipyramid (TBP) (τ = 0.50 for [1d]2,
where τ = 0.00 for a perfect SP and τ = 1.00 for a TBP geometry).16 These structures show that the new
ligands we prepared can accommodate both 5ꢀ and 6ꢀcoordinate binding motives, which is critical for
the mechanism we described above.
12

Page 13 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
Figure 9. Displacement ellipsoid (50%) representations of (a) [1c(THF)]2 and (b) [1d]2. Hydrogen
atoms and a coꢀcrystallized solvent molecule are omitted for clarity. See Supporting Information for
details.
Table 1. Selected Bond Distances and Angles for [1c(THF)]2 and [1d]2.
Fe–N1
(Å)
Fe–O1
(Å)
Fe–O2
(Å)
Fe–O2’
(Å)
Fe–O3
(Å)
[1c(THF)]2 2.070(1)
1.911(1)
1.903(1)
2.077(1)
2.063(1)
2.020(1)
2.012(1)
1.909(1)
1.909(1)
2.068(1)
[1d]2
2.057(2)
1.887(2)
1.977(2)
2.029(2)
1.845(2)
τ
N1–Fe–O
O1–Fe–O2
O1–Fe–O3
(°)
O2–Fe–O3
(°)
(°)
(°)
[1c(THF)]2
[1d]2
–
164.55(5)
163.91(5)
167.56(5)
167.21(5)
98.37(5)
97.24(5)
93.18(5)
94.67(5)
0.50
161.45(8)
131.25(8)
121.48(9)
106.90(8)
Kleij et al. reported that both monomeric and dimeric complexes are accessible with orthoꢀ
substituted N1–O3 binding ligands, where the monomeric form is more effective for coupling CO2 and
epoxides.10d Thus, it appears that forming the mononuclear species from the dimers is viable at normal
13

Catalysis Science & Technology
Page 14 of 22
View Article Online
DOI: 10.1039/C7CY01435J
reaction conditions. The same metalꢀoxygen coordination is also expected for the five coordinate species
[1d]2. The bridgehead oxygenꢀmetal bond can easily be cleaved to form the monomeric adduct in
solution, which will generate a five coordinate intermediate species that was used in our computer
modeling study described above.
Table 2. Selective formation of Cyclic Carbonates from the coupling of CO2 and Internal Epoxides.
Catalyst
Conv (%)
Selectivity (12g : 12g')
31
82
83
80
84
24
>50:1
4:1
1a
1b
1c
1d
1e
1f
23:1
49:1
4.5:1
>50:1
Experimental Confirmation. In order to verify the catalytic activity of the new Fe complexes
towards cyclic carbonate formation, we attempted the coupling reaction between CO2 and cyclohexene
oxide (11g) with NBu4Br as a coꢀcatalyst and the results are summarized in Table 2. The reaction
required high temperature (100 oC), high CO2 pressure (10 bar) and prolonged reaction time (2 ~ 8 h) to
obtain high conversion (>80%), which is in good agreement with the relatively high barriers that we
found in our calculations. Interestingly, as the reaction proceeds to full conversion, catalyst 1a produces
very little amount of polycarbonates showing a 12g:12g' product ratio of 50:1 in a low conversion rate
14

Page 15 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
of only 31%, as discussed above. In contrast, 1b produced a significant amount of polycarbonates, in
good agreement with the DFTꢀcalculated energy profiles, and we found a product ratio of 4:1, but in
much higher reactant conversion rate of 82%. Remarkably, catalyst 1c was confirmed to be much more
selective towards producing 12g in 83% yield and a product ratio of 23:1. By slightly modifying the
catalyst by adding a nitro functionality to the ligand, as shown in Table 2, we obtained 1d, which was
found to be the most selective system giving the product cyclized carbonate product 12g in 80% yield
with a ratio of 49:1. In order to gauge the scope of the reaction, catalyst 1d was used to prepare a variety
of cyclic carbonates (12g-12o) by the coupling with internal epoxides and CO2 in good isolated yield of
77ꢀ99% as shown in Table 3. These results qualitatively confirm our computational results, but also
highlight the limitations of DFTꢀbased predictive models. If the reaction energies discussed above were
taken by face value, we should be unable to detect any polymeric products with catalyst 1c and 1d, since
the energy difference of ~8 kcal/mol is much too great to be in quantitative agreement with the
experimentally determined ratio of 23:1 or 49:1. Similarly, our computer model does not precisely
predict that the catalysts 1a and 1b would be not selective for the cyclic carbonate formation per se – as
we discussed above, the key insight of the computer models is that the cyclic carbonate formation has a
much lower barrier when catalyst 1c is used for reasons that are plausible and easy to understand, while
the polymerization pathway is less impacted by the ligand modifications. The numerical results of the
simulations confirm this conceptual model, but the actual numbers are not in perfect agreement with the
experimental finding.
Obviously, there are computational inaccuracies and possibly other mechanisms that we may not
have been able to capture. Nonetheless, the qualitatively correct prediction and the notably improved
catalytic performance of the new system provides some justification and validates our approach of
computerꢀaided catalyst design. This work demonstrates therefore, that currently available
computational reaction modeling methods are not yet ready to be used as fully quantitative and accurate
models, but they are very much capable of delivering a plausible conceptual strategy with some
numerical results that can be augmented with specific experimental benchmark results. In this case, such
a strategy led to a significant improvement of the chemoselectivity and efficiency of the cyclic carbonate
forming catalysis based on a very plausible mechanistic model.
Table 3. Selective Formation of cyclic carbonates from the coupling of CO2 and and different
cyclohexene oxide derivatives.
15

Catalysis Science & Technology
Page 16 of 22
View Article Online
DOI: 10.1039/C7CY01435J
Entry Epoxide
Carbonate
1d/NBu4Br Time Yield
(mol%)
(h) (%)a
1
0.2 / 5.0
12
91
2
3
0.2 / 5.0
0.2 / 5.0
18
18
90
77
4
5
0.2 / 5.0
12
99
99
0.4 / 10.0 12
6
0.1 / 1.5
12
94
7
8
0.1 / 1.5
0.5 / 5.0
12
12
94
98
9
2.0 / 10.0 12
84
aIsolated yield
Computational Details. All calculations were carried out using DFT17 as implemented in the
Jaguar 8.9 suite of ab initio quantum chemistry programs.18 Structures were optimized with the
B3LYP19ꢀ23 functional including Grimme’s D3 dispersion correction24 and the 6ꢀ31G** basis set. The
Los Alamos LACVP25ꢀ27 basis that includes effective core potentials was used to represent Fe.
Additional single point calculations were performed on each optimized geometry using Dunning’s
16

Page 17 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
correlation consistent tripleꢀζ basis set ccꢀpVTZ(ꢀf)28 that includes a double set of polarization functions
to improve the electronic energy. For Fe, we used a modified version of LACVP, designated as
LACV3P, in which the exponents were decontracted to match the effective core potential with tripleꢀζ
quality. Solvation energies were evaluated by a selfꢀconsistent reaction field (SCRF)29ꢀ32 approach based
on accurate numerical solutions of the PoissonꢀBoltzmann equation. Solvation calculations were carried
out with the 6ꢀ31G**/LACVP basis at the optimized gasꢀphase geometry employing the dielectric
constants of ε = 15 for cyclohexanol. As is the case for all continuum models, the solvation energies are
subject to empirical parameterization of the atomic radii that are used to generate the solute surface. We
employed the standard set of optimized radii in Jaguar for H (1.150 Å), C (1.900 Å), P (2.074 Å), O
(1.600Å), and Rh (1.464Å). Analytical vibrational frequencies within the harmonic approximation were
computed with the 6ꢀ31G**/LACVP basis to confirm that all intermediates had no imaginary
frequencies except for transition states, which possessed one imaginary frequency.
Conclusions
Our studies suggest that the inefficiency of the Feꢀsalen complex as a catalyst for the formation
of cyclic carbonates using CO2 and epoxides is the result of the fourꢀcoordinate equatorial binding motif
of the salen ligand, leading to low conversion rates. Because there is no binding site available in the
equatorial positions, the CO2 cannot engage the metal properly when approaching the axially bound
epoxide and must do so in an outersphere mechanism. In this mechanism, the newly formed carbonate
must be rebound to the metal, which we propose is inefficient and leads to the poor catalyst performance.
This problem may be addressed by reducing the number of equatorial ligands to three, thus allowing the
carbonate to remain bound to the metal center at all times. The recently prepared catalyst by Kleij shows
such behavior. Our DFT studies show, however, that the barriers for the epoxide ringꢀopening and
carbonate cyclization steps are identical to the Feꢀsalen system. The structures obtained in our DFT
studies suggested that the reactivity may be enhanced by increasing the transꢀeffect, thus, making the
amino ligand in the axial transꢀposition to the epoxide a stronger Lewis base. To do so, we considered a
slight modification of the Nꢀdonor ligand to afford an imido ligand. The initial design principle was that
a more electronꢀrich metal center would also discourage the binding of a second epoxide to the metal
center, which we had initially assumed to be key to ringꢀopening polymerization. The imido ligand
showed very promising reaction energy profiles in our DFTꢀcalculations and much lower barriers were
obtained for the cyclic carbonate forming steps. Examining the polymerization mechanism, we found
surprisingly that the innersphere mechanism where both the carbonate and the incoming epoxide
substrate remain bound to the metal center is not feasible energetically. Instead, the carbonate must be
17

Catalysis Science & Technology
Page 18 of 22
View Article Online
DOI: 10.1039/C7CY01435J
detached first such that it can carry out the SN2 attack in an angle that is impossible in an innersphere
mechanism. This mechanistic insight leads to the conclusion that the ringꢀopening polymerization
reaction should be fairly insensitive to ligand modifications of the iron catalyst. In this case, the new
imido ligand is predicted to be chemoselective, since the barrier for the cyclic carbonate formation is
significantly lowered, while the intrinsically high barrier for the polymerization pathway remains
unaffected. These predictions were supported by subsequent experimental work, where we prepared the
proposed catalyst and found that the cyclic carbonate formation was indeed much improved and was
found to be highly selective over the polymerization pathway. The ability of this new catalyst in
selectively and efficiently preparing a variety of cyclic carbonate with internal epoxides was
demonstrated.
Associated Content
Supporting Information
Experimental procedure and characterization of compounds (1H and 13CꢀNMR spectra), Cartesian
coordinates of DFTꢀoptimized structures and computed energy component. This material is available
free of charge via the Internet at http://pubs.rsc.org/
Author Information
Corresponding Author
mbaik2805@kaist.ac.kr; yunholee@kaist.ac.kr; hwkim@kaist.edu
Author Contributions
ǁThese authors contributed equally to this work.
Notes
The authors declare no competing financial interest.
Acknowledgment
This work supported by the Saudi AramcoꢀKAIST CO2 Management Center and the Institute for Basic
Science (IBSꢀR10ꢀD1) in Korea.
References
(1) (a) Liu, Q.; Wu, L.; Jackstell, R.; Beller, M. Nat. Commun. 2015, 6, 5933. (b) Yu, B.; He, Lꢀ
N. ChemSusChem 2015, 8, 52ꢀ62. (c) Aresta, M.; Dibenedetto, A.; Angelini, A. Chem. Rev. 2014, 114,
1709ꢀ1742. (d) Appel, A. M.; Bercaw, J. E.; Bocarsly, A. B.; Dobbek, H.; DuBois, D. L.; Dupuis, M.;
18

Page 19 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
Ferry, J. G.; Fujita, E.; Hille, R.; Kenis, P. J. A.; Kerfeld, C. A.; Morris, R. H.; Peden, C. H. F.; Portis, A.
R.; Ragsdale, S. W.; Rauchfuss, T. B.; Reek, J. N. H.; Seefeldt, L. C.; Thauer, R. K.; Waldrop, G. L.
Chem. Rev. 2013, 113, 6621ꢀ6658. (e) Kielland, N.; Whiteoak, C. J.; Kleij, A. W. Adv. Synth. Catal.
2013, 355, 2115ꢀ2138. (f) New and Future Developments in Catalysis: Activation of Carbon Dioxide;
Suib, S. L., Ed.; Elsevier: Amsterdam, 2013. (g) Omae I. Coord. Chem. Rev. 2012, 256, 1384ꢀ1405. (h)
Carbon Dioxide as Chemical Feedstock; Aresta, M., Ed.; WileyꢀVCH: Weinheim, 2010. (i) Sakakura,
T.; Choi, J.ꢀC.; Yasuda, H. Chem. Rev. 2007, 107, 2365ꢀ2387.
(2) (a) Martín C.; Fiorani, G.; Kleij, A. W. ACS Catal. 2015, 5, 1353ꢀ1370. (b) Lu, X.ꢀB.; Ren,
W.ꢀM.; Wu, G.ꢀP. Acc. Chem. Res. 2012, 45, 1721ꢀ1735. (c) Lu, X.ꢀB.; Darensbourg, D. J., Chem. Soc.
Rev. 2012, 41, 1462ꢀ1484. (d) DeCortes, A.; Castilla, A. M.; Kleij. A. W. A. Angew. Chem. Int. Ed.
2010, 49, 9822ꢀ9837. (e) Sakakura, T.; Kohno, K. Chem. Commun. 2009, 1312ꢀ1330. (f) Darensbourg,
D. J. Chem. Rev. 2007, 107, 2388ꢀ2410.
(3) (a) Fiorani, G.; Guo, W.; Kleij, A. W. Green Chem. 2015, 17, 1375ꢀ1389. (b) Cokaja, M.;
Wilhelm, M. E.; Anthofer, M. H.; Herrmann, W. A.; Kühn, F. E. ChemSusChem 2015, 8, 2436ꢀ2454. (c)
Xu, B.ꢀH.; Wang, J.ꢀQ.; Sun, J.; Huang, Y.; Zhang J.ꢀP.; Zhang, X.ꢀP. Zhang, S.ꢀJ. Green Chem. 2015,
17, 108ꢀ122.
(4) For recent examples, see: (a) Zhang, H.; Lin, X.; Chin, S.; Grinstaff, M. W. J. Am. Chem. Soc.
2015, 137, 12660ꢀ12666. (b) Jia, F.; Chen, X.; Zheng, Y.; Qin, Y.; Tao, Y.; Wang, X. Chem. Commun.
2015, 51, 8504ꢀ8507. (c) Darensbourg, D. J.; Chung, W.ꢀC.; Yeung, A. D.; Luna, M. Macromolecules
2015, 48, 1679ꢀ1687. (d) Wang, Y.; Qin, Y.; Wang, X.; Wang, F. ACS Catal. 2015, 5, 393ꢀ396. (e)
Ohkawara, T.; Suzuki, K.; Nakano, K.; Mori, S.; Nozaki, K. J. Am. Chem. Soc. 2014, 136, 10728ꢀ10735.
(f) Wu, G.ꢀP.; Wei, S.ꢀH.; Ren, W.ꢀM.; Lu, X.ꢀB.; Xu, T.ꢀQ.; Darensbourg, D. J. J. Am. Chem. Soc.
2011, 133, 15191ꢀ15199.
(5) (a) Darensbourg, D. J.; Yarbrough, J. C. J. Am. Chem. Soc. 2002, 124, 6335ꢀ6342. (b)
Paddock, R. L.; Nguyen, S. T. J. Am. Chem. Soc. 2001, 123, 11498ꢀ11499. (c) Jacobsen, E. N.;
Tokunaga, M.; Larrow, J. F. PCT Int. Appl. WO 00/09463, 2000
(6) (a) Coates, G. W.; Moore, D. R. Angew. Chem., Int. Ed. 2004, 43, 6618ꢀ6639. (b) Qin, Z.;
Thomas, C. M.; Lee, S.; Coates, G. W. Angew. Chem., Int. Ed. 2003, 42, 5484ꢀ5487.
(7) Clegg, W.; Harrington, R. W.; North, M.; Pasquale, R. Chem. – Eur. J. 2010, 16, 6828ꢀ6843.
19

Catalysis Science & Technology
Page 20 of 22
View Article Online
DOI: 10.1039/C7CY01435J
(8) S, S.; Min, J. K.; Seong, J, E.; Na, S. J.; Lee, B. Y. Angew. Chem., Int. Ed. 2008, 47, 7306ꢀ
7309.
(9) (a) Fiorani, G.; Stuck, M.; Martín, C.; Belmonte, M. M.; Martin, E.; EscuderoꢀAdán E. C;
Kleij, A. W. ChemSusChem 2016, 9, 1304ꢀ1311. (b) Whiteoak, C. J.; Kielland, N.; Laserna, V.;
EscuderoꢀAdán E. C.; Martin, E.; Kleij, A. W. J. Am. Chem. Soc. 2013, 135, 1228ꢀ1231.
(10) (a) Whiteoak, C. J.; Martin, E.; EscuderoꢀAdán E.; Kleij, A. W. Adv. Synth. Catal. 2013,
355, 2233ꢀ2239. (b) Taherimehr, M.; AlꢀAmsyar, S. M.; Whiteoak, C. J.; Kleij, A. W.; Pescarmona, P. P.
Green Chem. 2013, 15, 3083ꢀ3090. (c) Whiteoak, C. J.; Martin, E.; Belmonte, M. M.; BenetꢀBuchholz,
J.; Kleij, A. W. Adv. Synth. Catal. 2012, 354, 469ꢀ476. (d) Whiteoak, C. J.; Gjoka, B.; Martin, E.;
Belmonte, M. M.; EscuderoꢀAdán E. C.; Zonta, C.; Licini, G.; Kleij, A. W. Inorg. Chem. 2012, 51,
10639ꢀ10649.
(11) (a) Witeoak, C. J.; Kielland, N.; Laserna, V.; CastroꢀGómez, F.; Martin, E.; EscuderoꢀAdán
E. C.; Bo, C.; Kleij, A. W. Chem. – Eur. J. 2014, 20, 2264ꢀ2275. (b) Darensbourg, D. J.; Yeung A. d.
Polym. Chem. 2014, 5, 3949. (c) Adhikari, D.; Nguyen, S. T.; Baik, B.ꢀH. Chem. Commun. 2014, 50,
2676ꢀ2678.
(12) Luinstra, G. A.; Haas, G. R.; Molnar, F.; Bernhart, V.; Eberhardt, R.; Rieger, B. Chem. –
Eur. J. 2005, 11, 6298ꢀ6314
(13) Gilli, P., Bertolasi, V., Ferretti, V., Gilli, G. J. Am. Chem. Soc. 2000, 122, 10405ꢀ10417.
(14) (a) Kim, H.; So, S. M.; Chin, J.; Kim, B. M. Aldrichimica Acta 2008, 47, 77ꢀ88. (b) Kim, H.;
Nguyen, Y.; Yen, C. P.ꢀH.; Chagal, L.; Lough, A. J.; Kim, B. M.; Chin, J. J. Am. Chem. Soc. 2008, 130,
12184ꢀ12191.
(15) (a) Tong, L. H.; Wong, Y.ꢀL.; Lee, H. K.; Dilworth, J. R. Inorg. Chim. Acta 2012, 383, 91ꢀ
97. (b) Zhang, J.; Liu A.; Pan, X.; Yao, L.; Wang, L.; Fang, J.; Wu, J. Inorg, Chem. 2011, 50, 9564ꢀ9570.
(c) Johnson, A. L.; Davidson, M. G.; Pérez Y.; Jones, M. D.; Merle, N.; Raithby, P. R.; Richards, S. P.
Dalton Trans. 2009, 5551ꢀ5558. (d) Su, W.; Kim, Y.; Ellern, A.; Guzei, L. A.; Verkade, J. G. J. Am.
Chem. Soc. 2006, 128, 13727ꢀ13735.
(16) Addison, A. W.; Rao, T. N.; Reedijk, J.; Vanrijn, J.; Verschoor, G. C. J. Chem. Soc., Dalton
Trans. 1984, 1349–1356.
20

Page 21 of 22
Catalysis Science & Technology
View Article Online
DOI: 10.1039/C7CY01435J
(17) Parr, R. G.; Yang. W., Density Functional Theory of Atoms and Molecules. Oxford
University Press: New York, 1989.
(18) Jaguar 8.9, Schrödinger, Inc., New York, NY, 2013.
(19) Slater, J.C., Quantum Theory of Molecules and Solids, Vol. 4: The SelfꢀConsistent Field
for Molecules and Solids. McGrawꢀHill: New York, 1974.
(20) Vosko, S. H.; WIlk, L.; Nusair, M., Can. J. Phys. 1980, 58, 1200.
(21) Becke, A. D., Phys. Rev. A 1988, 38, 3098.
(22) Becke, A. D., J. Chem. Phys. 1993, 98, 5648.
(23) Lee, C.; Yang, W.; Parr, R. G., Phys. Rev. B 1988, 37, 785.
(24) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. J. Chem. Phys. 2010, 132, 154104.
(25)
(26)
(27)
(28)
(29)
Hay, P. J.; Wadt, W. R., J. Chem. Phys. 1985, 82, 270.
Hay, P. J.; Wadt, W. R., J. Chem. Phys. 1985, 82, 299.
Wadt, W. R.; Hay, P. J., J. Chem. Phys. 1985, 82, 284.
Dunning, T. H., Jr., J. Chem. Phys. 1989, 90, 1007.
Marten, B.; Kim, K.; Cortis, C.; Friesner, R. A.; Murphy, R. B.; Ringnalda, M. N.;
Sitkoff, D.; Honig, B., J. Phys. Chem. 1996, 100, 11775.
(30)
Friedrichs, M.; Zhou, R.; Edinger, S. R.; Friesner, R. A., J. Phys. Chem. B 1999, 103,
3057.
21

Catalysis Science & Technology
Page 22 of 22
View Article Online
DOI: 10.1039/C7CY01435J
(31)
Edinger, S. R.; Cortis, C.; Shenkin, P. S.; Friesner, R. A., J. Phys. Chem. B 1997, 101,
1190.
22