At t MAX ¼ 10
À2 s, 262 was contracted to the superstate whose representative
structure was 22, while 796 was contracted to the one with 745 being a representative structure. Moreover, TS1270/1167 was obtained as the TS of the bottleneck
between the superstates whose representative structures were 745 and 1425. When
down to t MAX ¼ 10
À5 s, 796 was contracted to the superstate whose representative
structure was 116, resulting in TS511/541 as the TS of the bottleneck between the
superstates whose representative structures are 116 and 745. At t MAX ¼ 10
À6 s, 1167
was contracted to the superstate whose representative structure was 1167 itself, and
TS1282/1165 was obtained as the TS of the bottleneck between the superstates
whose representative structures were 1167 and 1425. At t MAX ¼ 10
À8 s, 1270 was
contracted to the superstate whose representative structure was 226, and TS80/350
was obtained as the TS of the bottleneck between the superstates whose representative structures were 745 and 226. At t MAX ¼ 10
À9 s, it was found that 745 was a
resting state because both 541 and 80 were contracted to the superstate whose
representative structure was 720. Further, TS81/166 was obtained as the TS of the
bottleneck between the superstates whose representative structures were 720 and
226. At t MAX ¼ 10
À10 s, 796 was contracted to the superstate whose representative
structure was 23, and 1270 was contracted to the superstate whose representative
structure was 1270 itself. Also, TS203/279 and TS475/1270 were obtained as the TS
of the bottleneck between the superstates whose representative structures were 23
and 116 and the ones with 226 and 1270 being representative structures, respectively. Additionally, because Fig. 5 shows a hybrid network consisting of approximate paths by the LUP method and IRC paths, the TS of the bottleneck obtained by
the contraction at t MAX < 10
À3 s was gained by reoptimizing the approximate paths
of the resultant bottlenecks.
The reaction mechanism extracted from the reaction path network by the RCMC
method is consistent with the reaction mechanism of the hydroformylation reported
so far. This study extracted the steps that took a longer timescale than 10
À10 s as
kinetically important steps. We could go on further; the CO coordination step to
CH 3 CH 2 -Co(CO) 3 , the H 2 coordination step to CH 3 CH 2 CO- Co(CO) 3 , and the H-H
bond dissociation step on Co, which proceed on a shorter timescale than 10
À10 s
(through barriers not exceeding ~22 kJ/mol), can be extracted by performing
contraction at t MAX < 10
À10 s. The above result shows that it was possible to execute
from the construction of the reaction path network by the SC-AFIR method to the
extraction of the reaction mechanism by the RCMC method without any human
intervention.
In addition to the main product propionaldehyde, the network in Fig. 5 suggests
the generation of other products, such as propan-1-ol by adding H 2 to 226 and propyl
formate by adding H 2 to 1425. Previous theoretical studies suggested the generation
of ethane [47, 83, 84]. Although paths giving ethane were obtained in the search,
little ethane was generated at 400 K on this reaction path network; rather, it was
found that ethane became the main product at 500 K. Besides this, many minor
byproduct candidates are predicted on this reaction path network. In an actual
reaction system, however, it can be imagined that 226 and 1425 are hardly generated
because of the rapid replacement of propionaldehyde in 720 and 745 with
74
M. Hatanaka et al.
À2 s, 262 was contracted to the superstate whose representative
structure was 22, while 796 was contracted to the one with 745 being a representative structure. Moreover, TS1270/1167 was obtained as the TS of the bottleneck
between the superstates whose representative structures were 745 and 1425. When
down to t MAX ¼ 10
À5 s, 796 was contracted to the superstate whose representative
structure was 116, resulting in TS511/541 as the TS of the bottleneck between the
superstates whose representative structures are 116 and 745. At t MAX ¼ 10
À6 s, 1167
was contracted to the superstate whose representative structure was 1167 itself, and
TS1282/1165 was obtained as the TS of the bottleneck between the superstates
whose representative structures were 1167 and 1425. At t MAX ¼ 10
À8 s, 1270 was
contracted to the superstate whose representative structure was 226, and TS80/350
was obtained as the TS of the bottleneck between the superstates whose representative structures were 745 and 226. At t MAX ¼ 10
À9 s, it was found that 745 was a
resting state because both 541 and 80 were contracted to the superstate whose
representative structure was 720. Further, TS81/166 was obtained as the TS of the
bottleneck between the superstates whose representative structures were 720 and
226. At t MAX ¼ 10
À10 s, 796 was contracted to the superstate whose representative
structure was 23, and 1270 was contracted to the superstate whose representative
structure was 1270 itself. Also, TS203/279 and TS475/1270 were obtained as the TS
of the bottleneck between the superstates whose representative structures were 23
and 116 and the ones with 226 and 1270 being representative structures, respectively. Additionally, because Fig. 5 shows a hybrid network consisting of approximate paths by the LUP method and IRC paths, the TS of the bottleneck obtained by
the contraction at t MAX < 10
À3 s was gained by reoptimizing the approximate paths
of the resultant bottlenecks.
The reaction mechanism extracted from the reaction path network by the RCMC
method is consistent with the reaction mechanism of the hydroformylation reported
so far. This study extracted the steps that took a longer timescale than 10
À10 s as
kinetically important steps. We could go on further; the CO coordination step to
CH 3 CH 2 -Co(CO) 3 , the H 2 coordination step to CH 3 CH 2 CO- Co(CO) 3 , and the H-H
bond dissociation step on Co, which proceed on a shorter timescale than 10
À10 s
(through barriers not exceeding ~22 kJ/mol), can be extracted by performing
contraction at t MAX < 10
À10 s. The above result shows that it was possible to execute
from the construction of the reaction path network by the SC-AFIR method to the
extraction of the reaction mechanism by the RCMC method without any human
intervention.
In addition to the main product propionaldehyde, the network in Fig. 5 suggests
the generation of other products, such as propan-1-ol by adding H 2 to 226 and propyl
formate by adding H 2 to 1425. Previous theoretical studies suggested the generation
of ethane [47, 83, 84]. Although paths giving ethane were obtained in the search,
little ethane was generated at 400 K on this reaction path network; rather, it was
found that ethane became the main product at 500 K. Besides this, many minor
byproduct candidates are predicted on this reaction path network. In an actual
reaction system, however, it can be imagined that 226 and 1425 are hardly generated
because of the rapid replacement of propionaldehyde in 720 and 745 with
74
M. Hatanaka et al.