was a slow process that could be observed with
iDPC-STEM imaging. To quantitatively describe
the benzene desorption process, we normal-
ized the contrast of benzene spots in differ-
ent images according to the contrast of the
zeolite framework in the related profile anal-
ysis, given that we had been observing the
same area (fig. S11).
Werecordedthetimeatwhichthezeolite
framework reached adsorption saturation as
time zero; at the same time, we kept the tem-
perature constant and switched the atmo-
sphere to vacuum. Figure 2, A to F, shows
real-time images of straight channels through-
out the desorption process. As benzene was
released, the contrast of the benzene spots in
straight channels continuously decreased (ini-
tially fast and then more slowly). Most benzene
molecules were desorbed within the first 40 min
(Fig. 2H). Finally, most of benzene molecules
were desorbed after heating to 923 K. Further-
more, we compared the in situ STEM results
with the thermal gravimetric analysis (TGA)
results (fig. S12). The initial decline rate of the
TGA desorption process was faster because of
the contribution of benzene molecules ad-
sorbed on the outer surface. Finally, most of
the benzene was desorbed in a similar time.
As benzene was released, the channel geo-
metry changed with the number of adsorbed
benzene molecules. With the decrease in the
contrast of benzene spots, the degree of chan-
nel deformation (described as the aspect ratio
ofDmax/Dmin) gradually decreased, and the
shape returned from an ellipse to an approx-
imately round shape (Fig. 2G). The deformation
degree of the straight channels was positively
correlated with the number of adsorbed ben-
zene molecules in the straight channels; this
was further confirmed in saturated ZSM-5
zeolites at different temperatures (fig. S13).
This relation indicated that the host-guest
interaction was related to the number of
adsorbed molecules and would further affect
the shape of pores.
Conversely, the geometry of the opening
pores also affected the arrangement and orien-
tations of confined benzene molecules (Fig. 2I).
With benzene desorption, whereas the pore
shapes tended to resemble ellipses with a
smaller curvature, the dominant configuration
of benzene molecules changed in response
(from 52° to 46°) and the angle distribution
became wider, indicating a weakened host-
guest interaction or confinement to facilitate
molecular vibration and rotation. Thus, as the
adsorbed benzene molecules deformed the
opening pores, the deformed geometry of
the zeolite pores dictated the dominant con-
figurations of guest molecules in return. In
this regard, this observed alterable geometry
of zeolite channels indicated that host-guest
interactions in flexible frameworks are not
immutable. To better understand the different
predominant pathways at different loadings,
apart from the interaction between guest
molecules, the variation in geometry of the
opening pores needs to be considered when
we describe the entropy and enthalpy of ad-
sorption ( 45 , 46 ).Ab initio molecular dynamics simulations at
different loadings
To gain more insight into the subcell defor-
mation of the ZSM-5 zeolites, we performed ab
initio molecular dynamics (AIMD) simulations
and calculated the distributions of the cor-
responding lengths of both axes (Fig. 3, A to
D). AIMD simulation is a powerful tool for
in situ tracking of the dynamic changes of
chemical bonds (such as bond lengths and
angles) for zeolite catalysts at the picosecond
scale under working conditions ( 9 , 25 ). Both
the maximal (Dmax) and minimal (Dmin) pore
sizes of straight channels (Fig. 3A) with ad-
sorbed benzene molecules of various loadings
were detected at the experimental temper-
ature (473 K). The STEM imaging mode ob-
tained images by scanning point by point,
so it often took several minutes to form an
image and much longer than the time scale
of the atomic thermal motion in the frame-
work. However, AIMD simulations can pro-
vide more detailed information on zeolite
framework thermal motion at a picosecond
time scale.
In the absence of adsorbed benzene mole-
cules, the distributions ofDmaxandDminwere
similar, with peaks located at ~9.1 and ~9.0 Å,
respectively. Tracing the variances in pore
size over time revealed that the pore chan-
nels were highly flexible and tended to be
circular statistically, but that there were
alternating shifts ofDmaxandDmin(Fig. 3B)
suggestive of a thermally activated breathing
motion of the zeolite pores.494 29 APRIL 2022¥VOL 376 ISSUE 6592 science.orgSCIENCE
(^07891011)
5
10
15
20
(^25) Dmin(0 C 6 H 6 ) Dmax(0 C 6 H 6 )
Dmin(1 C 6 H 6 ) Dmax(1 C 6 H 6 )
Dmin(4 C 6 H 6 ) Dmax(4 C 6 H 6 )
)
%(
no
it
u
bir
ts
i
D
Pore size (Å)
3000 6000 9000 12000 15000
8
9
10
11
(^) e
zi
s
er
o
P
(Å)
Time step
Dmax
Dmin
MFI framework
3000 6000 9000 12000 15000
8
9
10
11
Time step
Dmax
MFI framework Dmin
1 C 6 H 6
3000 6000 9000 12000 15000
8
9
10
11
Time step
Dmax
MFI framework Dmin
4 C 6 H 6
α=135 ~153
0 5 10 15 20
1.55
1.60
1.65
1.70
1.75
1.80
1.85
d (Å)
Si-O distance
1
2
(^345678)
(^910)
11
1312
(^17161514)
20
(^1918)
MFI
MFI_1C 6 H 6
MFI_4C 6 H 6
0246810
120
130
140
150
160
170
180
MFI
MFI_1C 6 H 6
MFI_4C 6 H 6
α (
°)
Si-O-Si bond
1
(^234)
5
6
(^987)
10
0246810
90
100
110
120
130
140
150
β (°)
O-Si-O bond
1
2
3 4
5
6
7
(^98)
10
MFI
MFI_1C 6 H 6
MFI_4C 6 H 6
AB C D
EFG H
Fig. 3. Investigating the chemical nature of MFI framework flexibility
via ab initio molecular dynamics simulation.(A) Distributions of pore
size for benzene (C 6 H 6 ) molecule adsorption with different loadings as
calculated from AIMD simulations at 473 K. (BtoD) Pore size evolution
with time for 0 (B), 1 (C), and 4 (D) C 6 H 6 molecules adsorbed inside MFI
zeolite. (E) Schematic diagram of the tetrahedral SiO 4 linkage with three
structural defining parameters. (FtoH) Statistical results for the Si-O-Si
angle (F), O-Si-O angle (G), and Si-O bond length (H) indicating a loose
network composed of rigid tetrahedral SiO 4 and soft hinges at the
oxygen atoms. In the statistics of each parameter, the corresponding
number of variables is listed in the inserted schematic. See figs. S14 to S17
for details. Time step: 0.5 fs.
RESEARCH | RESEARCH ARTICLES