You are on page 1of 11

nature catalysis

Article https://doi.org/10.1038/s41929-023-00913-8

Operando magnetic resonance imaging


of product distributions within the pores
of catalyst pellets during Fischer–Tropsch
synthesis

Received: 6 June 2022 Qingyuan Zheng1, Jack Williams1, Léonard R. van Thiel1, Scott V. Elgersma1,
Mick D. Mantle1, Andrew J. Sederman    1, Timothy A. Baart    2,
Accepted: 5 January 2023
G. Leendert Bezemer    2, Constant M. Guédon    2 & Lynn F. Gladden    1 
Published online: xx xx xxxx

Check for updates The optimization of a heterogeneous catalytic process requires


characterization of the catalyst at industrially relevant conditions and
length scales. Here we use magnetic resonance imaging to gain insight into
the Fischer–Tropsch synthesis occurring in a pilot-scale fixed-bed reactor
operating at 220 °C and 37 bar, for three H2/CO feed ratios. The molecular
diffusion and carbon number of the hydrocarbon products are spatially
resolved within both the reactor and individual 1 wt% Ru/TiO2 catalyst
pellets. These data highlight the importance of mass transfer, in addition
to the nanoscale catalyst activity, on catalyst performance. In particular,
a start-up time of up to three weeks is required for the steady state to be
achieved in the catalyst pores. Further, the average carbon number
present in the pores can be as much as double that in the product wax.
The operando characterization of water and oxygenates present in the pores
is also achieved. The presence of a water-rich liquid at the pore surface
is confirmed.

In heterogeneous catalysis, the catalytic process is directly affected and the intrinsic catalytic activity of the catalytic material. Among
by the physical structure and chemical formulation of the catalyst, these techniques, magnetic resonance, including NMR and magnetic
the intrinsic chemical kinetics and mass transfer characteristics of resonance imaging (MRI), is of particular interest since it can probe
the reaction and the internal structure and operating conditions of optically opaque systems directly and can measure multiple properties,
the reactor. It therefore follows that to design more effective cata- including molecular diffusion10, flow fields11 and temperature12 within
lytic processes, greater understanding of what is occurring inside the reactor. To date, MRI studies have been limited by insufficient spa-
the catalyst pellet and at the length scale of the reactor itself during tial resolution to differentiate species inside (intra-pellet) and outside
process operation is required. Operando imaging techniques with (inter-pellet) the catalyst pellets, and by the need to achieve sufficient
chemical resolution have previously been applied to catalytic reac- spectral resolution such that species of interest can be differentiated
tors; in particular, X-ray1,2, Raman3,4, ultraviolet3, infrared4,5 and nuclear based on their chemical shifts in the acquired spectra. Examples in
magnetic resonance (NMR)6,7 spectroscopies have been used. Reviews which chemical resolution has been achieved within a single pellet
of operando studies of catalysis have been reported8,9. These studies include imaging of propene hydrogenation within a Pt/Al2O3 pellet13
predominantly focus on the nature of the chemistry of the active site and the esterification reaction of propionic acid and 1-butanol within

1
Department of Chemical Engineering and Biotechnology, University of Cambridge, Cambridge, UK. 2Shell Global Solutions International B.V.,
Amsterdam, Netherlands.  e-mail: lfg1@cam.ac.uk

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

a Gas feed lines e

0.08

TI 1 2 FR = 2
r.f. coil FR = 1
FR = 0.5

Mole fraction
Gradient set
control

control

control
Flow

Flow

Flow
TI
Cooling
jacket

3 GC 0
0 20 40 60 80 100
Carbon number NC
4 5 PC 6
CO

N2
H2

f
Wax products Water/light products
–4
collection vessel collection vessel

ln (mole fraction)
α = 0.94
b 40 c 100 d
–5
2.0
Outlet H2/CO
30 95
α = 0.93
SC5+ (%)
XCO (%)

20 90
1.0
–6
10 85 α = 0.88
0.5

0 80 0 30 35 40 45 50
2 1 0.5 2 2 1 0.5 2 2 1 0.5 2
Carbon number NC
FR FR FR

Fig. 1 | Schematic of the reactor and reaction data. a, Schematic of the from the reactor outlet. f, Anderson–Schulz–Flory plots for the outlet wax
pilot-scale FT reactor rig: 1, fixed-bed tubular reactor; 2, NMR spectrometer; products with NC in the range 30–50. The chain growth probability α obtained for
3, gas/liquid separator; 4, storage of heavy hydrocarbons; 5, storage of light each feed ratio is shown. For NC < 30, some gas-phase products have been formed
hydrocarbons and water; and 6, wet gas flow meter. The parts of the rig equipped and therefore the wax (liquid) product data in e will not reflect the total number
with trace heating are highlighted in red. TI and PC denote temperature indicator of species (gas and liquid) of that carbon number produced during the reaction
and pressure controller, respectively. r.f., radio frequency. b, CO conversion (XCO) (as predicted by the Anderson–Schulz–Flory model); hence, these data are not
at each feed ratio (FR). c, The selectivity (S) to C5+ species. d, The H2/CO ratio at included in the analysis.
the reactor outlet. e, Carbon number distributions of the wax products collected

a hydrogel bead14. However, the use of chemical shift methods in study- in what phase is water present and how does it exist within the pore
ing complex polymerization reactions such as the oligomerization of space20,22? Further, design of reaction operating conditions will often
light olefins15 and Fischer–Tropsch (FT) synthesis16 is limited because be achieved through numerical modelling or simulation, which relies
the reaction products consist of species from the same homologous on assumptions of the product distribution formed inside the catalyst
series and therefore their spectral resonances have similar chemical pores; such information has not yet been made under true operando
shifts; this problem is made even more challenging because of the conditions (that is, employing millimetre-scale FT catalyst pellets
spectral line broadening present when species are confined within the operating under industrially relevant conditions).
pores of an oxide-based catalyst. In this work, magnetic resonance techniques are applied to char-
The FT reaction is the central part of processes that pro- acterize the FT reaction occurring in a pilot-scale fixed-bed reactor.
duce synthetic fuels and chemicals from resources such as natu- The present work reports one-dimensional (1D), two-dimensional
ral gas17, biomass 18 and carbon dioxide 19. The low temperature (2D) and three-dimensional (3D) imaging of the distribution of liquid
operation of the FT reaction converts carbon monoxide and hydro- products within the reactor. Spatial resolution of the hydrocarbon
gen to primarily n-alkanes that have broad carbon number (NC) chain length is achieved using a recently developed approach in which
distributions16: spatial maps of the distributions of molecular diffusion coefficients,
D, of hydrocarbon molecules present inside the catalyst pellets are
nCO + (2n + 1) H2 → Cn H2n+2 + nH2 O. (1) measured, and are then transformed to the carbon number distri-
butions characterizing those species23,24. Consistency with 2D NMR
The FT reaction is a particularly challenging reaction to under- spectroscopy data is confirmed. The distributions of D were measured
stand. Much attention has necessarily focused on the intrinsic activity using spatially resolved pulsed field gradient NMR. Spatial resolution is
and selectivity of the catalyst. However, to optimize a real catalyst, achieved such that molecular diffusion and hence hydrocarbon chain
often in the form of millimetre-scale pellets, the catalyst conversion length can be discriminated inside and outside of the catalyst pellets.
will be influenced by more than the intrinsic reaction kinetics alone. While this paper focuses on the ability of MRI to monitor the evolution
Common questions that arise in studying FT include the following: of carbon number distribution in real time, spectroscopic data are also
How is intrinsic catalyst performance modified by including the active shown to characterize the presence of water and oxygenates inside
sites inside a catalyst pellet20,21? Water is formed during the reaction; the pores.

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

Results
Fischer–Tropsch reaction Inert packing
The FT reaction study was carried out in a pilot-scale reactor rig
designed to operate up to 250 °C and 51 bar. The reactor body is con-
structed of silicon nitride (Si3N4), a non-magnetic ceramic material, y Catalyst layer 1
and has an inner diameter of 20 mm. The reactor set-up inside the mag- x
netic resonance magnet is shown in Fig. 1a. The reactor was placed into z
the magnet of the NMR spectrometer with the catalyst bed located in 1.0
Inert packing
the centre of the radio frequency (r.f.) coil, where the NMR signal was
acquired. The catalyst is a 1 wt% Ru/TiO2 extrudate of diameter 2 mm
and length 5 mm. The catalyst pellets were packed into three layers in Catalyst layer 2
the reactor, with the height of each catalyst layer being ~3 mm. In total,

Intensity (a.u.)
2.7 g catalyst was used. The three catalyst layers were separated by
~10-mm-thick layers of non-porous, inert silicon carbide (SiC) particles Inert packing
of 0.3 mm in diameter to achieve isothermal operating conditions.
More details on the catalyst and the reactor set-up are described in
Catalyst layer 3
the Methods.
The reaction was started at a H2/CO molar feed ratio of 2 and then
reduced to a feed ratio of 1 and then 0.5. The H2/CO ratio was adjusted Inert packing

by changing the partial pressure of H2 while the partial pressure of 0.1

CO was maintained at 10.3 bar. After the measurements at feed ratio Glass wool
0.5, the feed ratio was reset to 2 to check the stability of the catalyst.
Figure 1b,c confirms that the catalytic conversion occurring in the
pilot-scale reactor behaves as expected. In particular, CO conver-
sion (Fig. 1b) decreases whereas C5+ selectivity (Fig. 1c) increases with
Flow
decreasing feed ratio, consistent with the literature25–27. The data shown
in Fig. 1d confirm that when the reactor is operating close to the stoi-
chiometric ratio of ~2.1 (this number depends on the exact product Fig. 2 | Imaging catalyst bed structure and liquid distribution within the
distribution), the H2/CO ratio remains constant along the length of reactor. The 3D 1H intensity image of the reactor acquired at TOS = 400 h and a
the reactor (that is, the outlet H2/CO ratio is equal to the inlet H2/CO H2/CO feed ratio of 2. The field-of-view = 25(x) × 25(y) × 80(z) mm3 and the spatial
ratio for a feed ratio of 2). As the feed ratio decreases from the stoi- resolution = 195(x) × 195(y) × 313(z) µm3. The intensities were normalized to the
chiometric ratio, the H2/CO ratio at the exit decreases from the feed maximum intensity of the image.
condition. Finally, the gas chromatography (GC) analysis of the waxes
(liquid hydrocarbon products) collected at the exit from the reactor
confirms that the carbon number distributions (Fig. 1e) of the waxes the CH4 and C5+ selectivity as measured by GC on the exit stream of the
become heavier and broader as the feed ratio decreases with a larger reactor. The data of Fig. 3b,d show that while the GC might indicate
chain growth probability at a lower feed ratio (Fig. 1f). These data are steady-state behaviour at a TOS of ~51 h, coinciding with the complete
consistent with the Anderson–Schulz–Flory model16,28. pore filling, the intra-pellet product distribution continues to change
until at least TOS = 150 h, when liquid saturation of the entire catalyst
Imaging of liquid accumulation within the reactor layer is achieved. Further, the magnetic resonance data of Fig. 3c reveal
Figure 2 shows 3D MRI 1H intensity images throughout the reactor. The that changes in liquid composition (characterized by NC) in the pores,
1
H signal intensity increases with an increase in the number of 1H species as a result of mass transfer processes, are such that the achievement
present within each image voxel and hence reflects the distribution of of a steady state in product composition within the pores requires
liquid products within the reactor. The three catalyst layers are clearly TOS > 500 h. Similar timescales for reaching the steady state are
identified, as is the presence of liquid product in the inert packing, with observed for the other catalyst layers. This suggests the reaction should
substantial accumulation of product towards the bottom of the bed. be operated for at least two to three weeks to stabilize before taking
Figure 3a shows four 2D 1H intensity images acquired from layer 1 heavy wax samples for characterization and is consistent with the work
of the catalyst bed during the reaction start-up at a feed ratio of 2. Up of Pinard et al.29, who reported variations in wax compositions up to
to a time-on-stream (TOS) of ~51 h, signal intensity is associated with 21 days after reactor start-up. This observation has important implica-
liquid-phase 1H-containing species from inside the catalyst pellets only. tions for how catalyst screening studies are performed. Finally, Fig. 3e
After this time, a further increase in signal intensity is associated with combines data from Fig. 3b–d to show how NCand C5+ selectivity change
the inter-pellet liquid only (Supplementary Fig. 1). Figure 3b shows as a function of pore saturation during start-up. The average carbon
a time series of the total integrated signal intensity acquired from number inside the catalyst pores decreases strongly as the liquid satu-
layer 1 over the first 200 h of reactor operation. Once the saturation ration of the pore space increases, thereby showing the potential for
of the pore space with liquid is complete, the liquid uptake in the layer controlling product selectivity through controlling the level of pore
increases more slowly up to TOS = 150 h, after which the liquid content saturation of the catalyst. The negative effect of pore saturation on the
in this section of the reactor plateaus. Since signal is seen only inside selectivity to heavy products can be attributed to the increasing extent
the pellets up to TOS = 51 h, it is possible to use this image to produce a of the diffusion limitation of CO and H2 as the pores are increasingly
mask (Supplementary Fig. 2), which is then used to identify pixels that saturated with wax. At the reaction conditions used, this leads to an
are unambiguously associated with the signal from the liquid product increase in H2/CO ratio towards the centre of the catalyst pellet20,30.
inside the catalyst pellets (that is, from the intra-pellet space). Analysis
of only the intra-pellet data is presented here. Spatially resolving molecular diffusivities and carbon
The average diffusion coefficient of the intra-pellet liquid in layer numbers
1 is plotted as a function of TOS in Fig. 3c, along with the average carbon The 2D masked images showing the average diffusion coefficient, D ,
number, NC, calculated from these data (Methods). Figure 3d shows of intra-pellet liquid are presented in Fig. 4. For each feed ratio, the

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

a b 1.2
TOS = 4.8 h 22.3 h 34.0 h 50.9 h
1.0

Intensity (a.u.)
0.8
Layer
0.6 saturation
Pore
0.4 saturation
0.2

0
0 Intensity (a.u.) 1 0 50 100 150 200
TOS (h)

c 8 55 d 6 94 e 91.5
50
7 50 5 93
D (10–10 m2 s–1)

6 4 92 91.0 45
45

SCH4 (%)

SC5+ (%)
SC5+ (%)
5 3 91

NC

NC
40
40
4 2 90 90.5
35
3 35 1 89

2 30 0 88 90.0 30
0 100 200 300 400 500 600 0 50 100 150 200 250 300 0.2 0.4 0.6 0.8 1.0

TOS (h) TOS (h) Pore saturation

Fig. 3 | Effect of pore filling and mass transfer processes on catalyst normalized to that acquired at TOS = 150 h. c, Average diffusion coefficient and
selectivity. a, The 2D 1H intensity images of catalyst layer 1 acquired during the corresponding carbon number of the intra-pellet liquid in catalyst layer 1
reaction start-up at a H2/CO feed ratio of 2 showing the filling of catalyst pores by measured during the reaction start-up. The standard errors of D and NC are
liquid products (field-of-view = 22(x) × 22(y) mm2, slice thickness = 3 mm and estimated as 0.02 × 10−10 m2 s−1 and 0.1, respectively, using the results acquired at
spatial resolution = 172(x) × 172(y) μm2). The signal intensity of each image was TOS > 500 h for six repeat measurements. d, The CH4 and C5+ selectivity (S)
normalized to the maximum intensity of the image acquired at TOS = 50.9 h. measured by GC during start-up. To highlight the initial variation, the data up to
b, Total signal intensity of catalyst layer 1 showing the liquid accumulation in TOS = 300 h are shown; the data at TOS > 300 h remain stable. e, Correlation
pores and in the inter-pellet space of the catalyst layer. The signal intensities were between the extent of pore saturation and catalyst selectivity.

FR = 2 FR = 1 FR = 0.5 molecular diffusion coefficient decreases down the bed, as the reaction
z
proceeds. The values of D for a given catalyst layer decrease with
y
decreasing feed ratio. Heterogeneity in the D distribution is observed
x throughout the catalyst layers and also within individual catalyst pel-
lets for all catalyst layers and feed ratios. While the present study
10
focuses on the intra-pellet liquid, the diffusion distribution of the
D (10–10 m2 s–1)

8
inter-pellet liquid was also analysed. As expected, the diffusion coef-
6 ficients of molecules in the inter-pellet space extend to larger values
4
than are found in the intra-pellet space due to the absence of tortuos-
ity constraints. The distributions of molecular diffusivity measured at
2
individual pixels in Fig. 4 were then converted to carbon number (NC)
1
distributions. The 2D maps of average carbon number, NC, are shown
in Fig. 5a. Heterogeneity in the product distribution is seen across the
Intensity (a.u.)

bed. Such heterogeneity at the top of the bed is likely due to the mald-
istribution of liquid accumulation at the entrance of the reactor
(Supplementary Fig. 3). Heterogeneity in NC is also seen within
individual catalyst pellets. The probability density functions of NC
0
obtained from the NC maps in Fig. 5a are shown in Fig. 5b. As expected
from the data shown in Fig. 5a, for all three catalyst layers, the probabil-
ity density functions of NC shift towards larger NC with decreasing
feed ratio25,27. Figure 5b highlights differences in catalyst behaviour
Fig. 4 | 2D maps of the average diffusion coefficient of intra-pellet products. along the length of the reactor for the different feed ratios. For exam-
At each image pixel, a distribution of molecular diffusion coefficient, D, was ple, the feed ratio of 2 shows little change in the product distribution
obtained using the spatially resolved pulsed field gradient NMR experiment, formed within the catalyst pellets down the reactor. However, as the
from which the average diffusion coefficient was calculated. The spatial feed ratio decreases, a shift to heavier products as reaction proceeds
resolution is 172(x) × 344(y) μm2; all other imaging parameters are the same as down the bed is observed. This is explained by the decrease of the
stated for Fig. 3a. To improve visualization, the D maps were first interpolated to H2/CO ratio down the reactor (Fig. 1d), which leads to heavier products
1,024 × 1,024 matrices and then superimposed on the intensity images of the
being produced in the lower part of the reactor. In addition to the effect
catalyst layers. In this way, the D values are represented by the colour map, while
of the H2/CO ratio, the enhanced readsorption of olefins20 and higher
the brightness of a pixel indicates local signal intensity. In the D maps, some
water partial pressure31 in the lower regions of the reactor are also likely
pixels at the edge of the pellets are subject to large error and therefore not shown
(error analysis in Methods). The rows from top to bottom show the results for
to contribute to the increasing average carbon number down the reac-
catalyst layers 1–3, and the columns show the results for different feed ratio (FR) tor. Figure 5b also shows broadening of the NC distributions as the feed
values. The D values have standard errors of 0.20 × 10−10 m2 s−1 as estimated using ratio decreases. For completeness, the NC determined by diffusion was
three repeat measurements. validated by comparing the results with those measured using an

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

a FR = 2 FR = 1 FR = 0.5 b
0.3
z y FR = 2
FR = 1
0.2 FR = 0.5
x

80 0.1
70

Probability distribution function


60
0.4
NC

50
40

Flow direction
0.3
30
0.2

1 0.1
Intensity (a.u.)

0.3

0.2

0
0.1

0
30 40 50 60 70
NC

Fig. 5 | Average carbon numbers of intra-pellet products. a, The 2D maps of NC The NC values have standard errors of ~1 as estimated using three repeat
of the intra-pellet products. For each pixel, a carbon number distribution was measurements. The grey box on the image for layer 3 acquired at feed ratio 0.5
obtained and NC was calculated. The resolution, display and interpretation of the identifies the local region from which the single pellet analysis shown in Fig. 6 is
images are the same as in Fig. 4. The rows from top to bottom show the results for performed.
catalyst layers 1–3. b, Histograms of the NC values shown in the images in a.

independent (not spatially resolved) method based on 2D NMR characterize the wax formed within the catalyst pores, might seem
correlated spectroscopy (COSY)32. Good agreement is obtained appropriate based on the collected wax data, but the wax actually
between the two methods to within experimental error; the NC of the present in the pores can differ by up to 100%. This has obvious implica-
whole catalyst bed is consistent with the average values of NC obtained tions for understanding the reaction conditions and modelling the
from the three catalyst layers. The 2D COSY data are reported in reaction. Importantly, the concentration of the reactants dissolved in
Supplementary Fig. 4. the liquid wax product, determined by their solubility and diffusion
Figure 6a shows an expanded image of the region identified in coefficients within the wax, is a function of the NC of the wax.
Fig. 5a to highlight the variation of carbon number within a single cata- Figure 7b,c shows diffusivity and solubility data estimated for H2 and
lyst pellet. The probability distribution functions of the NC obtained CO using waxes with an NC in the range 20–80. The molecular mobility
from the catalyst pellet highlighted in Fig. 6a are presented in Fig. 6b. (characterized by diffusion coefficient) and solubility (characterized
The probability distribution functions of NC in the individual pellet by Henry’s law constant35) of the two gases in the waxes vary notably
show similar variation with feed ratio to that observed for the entire as a function of carbon number. In particular, using n-C28 as the repre-
catalyst layer (Fig. 5b). This confirms that a heterogeneity of reaction sentative wax in the pore space of the catalyst overestimates diffusivi-
occurs at a length scale smaller than the size of individual pellets. ties and the Henry’s law constants considerably. The magnitude of the
Similar data from a second pellet are shown in Supplementary Fig. 5. differences between what is estimated based on exit composition rela-
The 2D images shown in Figs. 4–6 represent the catalyst behaviour tive to actual wax composition within the pores will lead to substantial
averaged over a 3 mm slice thickness along the length of the reactor errors, which are summarized as follows. Using n-C28 as the solvent
(Methods). The level of information obtained and the quality of the overestimates the diffusivities of H2 and CO by 7–18% and 11–33%,
data are further demonstrated by Fig. 6c and Supplementary Fig. 6, in respectively, with the error becoming larger at lower feed ratios, when
which the data from a single pixel are plotted. As expected, the carbon heavier products are produced. As a result, the extent of the diffusion
number distribution of liquid from the selected pixel becomes heavier limitation of H2 and CO within the pellets will be underestimated to an
and broader as feed ratio decreases. increasing extent as feed ratio decreases. Importantly, the extent to
Figures 4–6 show that it is now possible to measure the molecular which the CO diffusivity is overestimated is greater than that for H2,
mobility directly and hence estimate the carbon number of the prod- which will change the H2/CO ratio present in the catalyst pellets from
ucts within the catalyst pellets during the FT reaction. Figure 7 explores that expected based on the feed condition; this will affect catalyst
further the importance of understanding product distributions in performance20,30. For the solubility data in Fig. 7c, the Henry’s law
catalyst pellets and how the insights provided by MRI yield information constants calculated using an NC of 28 overestimate the true values
that provides valuable new data on which to base numerical models of for H2 and CO by 15–98% and 13–80%, respectively. This suggests that
FT catalysis in the reactor environment. Figure 7a highlights the the actual mole fractions of H2 and CO in pores are respectively 1.2–2
difference in wax (that is, hydrocarbon mixture) composition inside times and 1.1–1.8 times larger than those commonly estimated, which
the catalyst pellets and in the collected wax at the reactor exit. Measure- will influence the reaction rate and the local H2/CO ratio. In addition to
ments of the intra-pellet NC are plotted, along with the results of the the impact on the behaviour of H2 and CO, the composition of
collected wax as a function of H2/CO feed ratio. These data confirm that intra-pellet products also affects the diffusion and solubility of indi-
an NC of 28, which is commonly used in numerical modelling33,34 to vidual hydrocarbon species. In particular, the diffusion and solubility

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

a b 0.3 c 0.07
FR = 2 FR = 2
80 FR = 1

Mole fraction (a.u.)


Probability density
FR = 1

function (a.u.)
0.2 FR = 0.5 FR = 0.5
70

60

NC
0.1
50

40
0 0
30 40 50 60 70 0 20 40 60 80 100 120

NC NC

Fig. 6 | Carbon number distributions within an individual catalyst pellet. a, the three feed ratios studied from the pixel identified by the arrow in
Zoomed-in map of NC from catalyst layer 3 acquired at feed ratio of 0.5 (shown in a. The shadow regions in c indicate the standard error (s.e.) of the distributions
Fig. 5a). b, Histograms of NC from within the highlighted pellet in a (purple box), estimated using three repeat measurements (n = 3). The result at each carbon
for the three feed ratios studied. c, Distributions of carbon number for each of number is presented as the mean value ± s.e.

of olefins have been reported to affect the readsorption of these species Conclusions
on the catalyst surface and hence affect catalyst selectivity26. Of course, This work reports operando imaging of the product distribution
the spatial heterogeneity in the NC within pellets and catalyst layers formed within individual catalyst pellets during FT synthesis in a
shown in Figs. 5 and 6 will also lead to further heterogeneity in the dif- pilot fixed-bed reactor; the work is performed at industrial reaction
fusion and solubility of reactants and products. conditions and in catalyst pellets and a reactor of an industrially
Ex situ measurements of intra-pellet product distribution have relevant scale. These data demonstrate the importance of oper-
been conducted previously21,29. These have shown an NC of the ando studies at these length scales if the real catalytic process is to
intra-pellet products of ~36 (ref. 21) and ~45 (ref. 36) at feed ratios of 2 be understood. Such insights complement operando studies at the
and 1, respectively, for a cobalt catalyst, and ~37 at a feed ratio of 2 nanoscale, which focus more on the intrinsic catalytic activity of a
(ref. 29) for a cobalt–ruthenium catalyst. These results are consistent catalytic material. The significance of being able to study the cata-
with the operando data reported in this paper that were acquired at lyst working at the pellet and reactor scales is that the study includes
the same feed ratios despite using a different catalyst, and they rein- the effect of mass transfer processes on the overall catalytic process,
force the observation that the intra-pellet wax composition is signifi- which can substantially modify the catalytic performance from that
cantly different from that of the wax collected at the reactor outlet. which is expected from the feed composition and reactor operating
conditions.
Operando characterization of water in pores The data reported show two main results. First, the carbon number
To understand how the reaction proceeds inside the catalyst pores, it distribution within the pores of the catalyst decreases with an increase
is important to track not just hydrocarbon formation but also water in the saturation of the pores with hydrocarbon product. Second, while
production and the phase and environment in which it exists, because of GC data suggest a start-up time to steady-state behaviour over time-
its impact on catalyst activity31,37, selectivity31 and stability37,38. Figure 8 scales of ~51 h, the magnetic resonance data show that this is actually
shows a 2D spin–lattice relaxation time versus chemical shift spectrum the timescale over which saturation of the pore space is occurring,
in which 1H-containing species that share overlapping chemical shift and that liquid distribution throughout the reactor takes considerably
signals are spectrally resolved through their nuclear spin–lattice (T1) longer to reach a steady state. Within the particular layer of the catalyst
relaxation times, which have been shown to be an indicator of sur- bed reported, the molecular diffusivity of the hydrocarbon products
face interaction39,40. Figure 8 confirms that oxygenates and water are is still evolving at ~500 h TOS. During this time, the product distribu-
formed along with hydrocarbons. The chemical shift and T1 relaxation tion is changing and hence so will the diffusivity of the H2 and CO gases
time of the water signal provide insight into the phase of the water. within the pore space, which are sensitive to the carbon number of the
The chemical shift of the water signal at 3–5 ppm identifies the water as wax in which they are diffusing. Further, the solubility of these gases
being in a water-rich liquid phase; gas-phase water and dilute water in in the wax inside the pores will change over these timescales, and the
the wax phase would have a chemical shift of ~0 ppm (ref. 39). Moreover H2/CO ratio will deviate significantly from that expected from the
the T1 value associated with the water of 10−3–10−2 s is characteristic feed condition.
of water interacting with the pore surface; water existing in the pore Spatially resolved maps of the molecular diffusion coefficients of
centre would have a T1 of ~1 s (ref. 39). The observation of a water-rich product hydrocarbon species within the pellets are reported, and from
layer covering the surface and coexisting with wax is consistent with the these, maps of average hydrocarbon chain length are calculated. These
predictions of molecular dynamics simulations for the case of hydro- data are consistent with the predictions of the Anderson–Schulz–Flory
philic pores22. This had not, until the present work, been confirmed mechanism: heavier hydrocarbon chains are produced at lower
under operando conditions, to the best of our knowledge. Using the H2/CO feed ratios. The product distribution is also seen to broaden with
signal intensity in the 2D spectra, the weight fraction of surface water in a decrease in feed ratio. Importantly, the data show that significantly
the intra-pellet liquid was calculated as 0.57 ± 0.05 wt%, 0.16 ± 0.07 wt% heavier hydrocarbon chains exist within the pore space of the catalyst
and 0.05 ± 0.04 wt% for feed ratios of 2, 1 and 0.5, respectively (error than are produced in the product wax collected at the exit of the reac-
analysis in Methods). The decrease of water composition is consistent tor. Such measurements made under operando conditions provide
with the decrease in CO conversion as feed ratio decreases. These meas- important data and insight in the context of more accurate model-
urements are currently being combined with the MRI modality to gain ling of the FT catalytic process. Operando 2D NMR spectroscopy also
far greater insight into the role of water during the FT process, along confirms the presence of water inside the reactor and that this water
with the characterization of oxygenate species. These studies will be exists in the form of a water-rich liquid phase at the surface of the pores
extended to catalysts of different formulation and physical structure. within the catalyst pellets.

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

a b Intra-pellet liquid c Intra-pellet liquid


70 1.6 0.6 600
Intra-pellet liquid H2 H2

Henry’s constant (bar)


De of CO (10–8 m2 s–1)
De of H2 (10–8 m2 s–1)
60 Collected wax 1.5 500
CO CO
0.5
50 1.4 400
NC

40 1.3 300
0.4
30 1.2 200

20 1.1 0.3 100


0.5 1.0 1.5 2.0 20 30 40 50 60 70 80 20 30 40 50 60 70 80
H2/CO feed ratio NC NC

Fig. 7 | Comparison between the intra-pellet liquid products and collected values. The values were calculated according to the method of Marano and
wax. a, NC of the intra-pellet products in the reactor averaged over all layers are Holder35. The black dashed vertical lines in b and c indicate the values calculated
compared with those of the wax collected at the reactor exit for the three feed for NC = 28, a value commonly used in the literature. The shaded areas in b and c
ratios studied. The NC values have standard errors of ~1 as estimated using three indicate the diffusion and solubility at the true composition of the intra-pellet
repeat measurements. b, Effective diffusivity, De, of H2 and CO dissolved in wax liquid as determined in this work. The lower and upper bounds of NC in the
mixtures of different NC values. The De values were calculated according to shaded areas are the NC values measured at catalyst layer 1 for FR = 2, and at
Zheng et al.48. c, Henry’s law constant of H2 and CO in wax mixtures of different NC catalyst layer 3 for FR = 0.5, respectively.

1 cylinders were mixed before entering the top of the reactor. The gas
Intensity (a.u.)

flow rates were controlled by mass flow meters calibrated to measure


flow rates in the range 0.5–30 NL h−1, where NL is a normal litre defined
as the mass of 1 litre of gas at a pressure of 1 atmosphere and a tem-
0 Population (a.u.) perature of 20 °C. The flow rate of the effluent gas was measured with
0 1 a calibrated wet gas flow meter. Trace heating was mounted along the
2
10
Hydrocarbons feed and effluent lines of the rig and on the external wall of the reactor
1 to provide heating as required. The temperature of the catalyst bed was
10
measured by two thermocouples located above and below the catalyst
bed inside the reactor; the temperature was controlled to an accuracy
T1 relaxation time (s)

0
10
of ±0.5 °C. The pressure of the reactor was controlled to an accuracy
–1
of ±0.1 bar with an Equilibar back pressure controller installed down-
10
Oxygenates stream of the reactor. The reactor rig was controlled by the National
–2
Instruments LabVIEW application. The wax products (mainly C10+)
10
were collected from the reactor outlet and analysed offline using GC.
–3 Water The gas-phase products (mainly C1–C5) and unreacted feed gases were
10
flowed to another GC instrument for online composition analysis.
–4
10
20 15 10 5 0 –5 –10 Experimental procedure
Chemical shift (ppm) Throughout the experiment, ramp rates of 1 °C min−1 and 0.3 bar min−1
were used for temperature and pressure operation, respectively. Prior
Fig. 8 | Operando characterization of water in catalyst pores. The 2D
1
H spectrum, correlating chemical shift and T1 relaxation time, is used to
to the FT reaction, the liquid–gas separators for wax products (vessels 3
discriminate hydrocarbons, oxygenates and water present in the reactor. The and 4 in Fig. 1a) were heated to 160 °C and 140 °C, respectively, and the
1D projections of the 2D spectrum onto the chemical shift and T1 axes are also liquid–gas separator for light products (vessel 5 in Fig. 1a) was cooled to
shown. The chemical shift is relative to the 1H resonance of tetramethylsilane. 3 °C. The gas feed and effluent lines were heated to 220 °C and 170 °C,
respectively. Before starting the reaction, the catalyst was reduced
in situ at 250 °C and 6 bar in a H2–N2 flow with a H2 flow of 1 NL h−1 and N2
flow of 20 NL h−1. The reduction took 40 h and was considered complete
Methods when there was no change in the H2 concentration in the effluent gas as
Materials measured by GC. After the catalyst reduction, the H2 flow was turned off
The catalyst, provided by Shell, has a Brunauer–Emmett–Teller surface and the N2 flow rate was reduced to 3 NL h−1. The reactor was cooled to
area of 44.8 m2 g−1, a Barrett–Joyner–Halenda average pore diameter 160 °C and depressurized to ambient pressure. After reaching 160 °C,
of 28.6 nm and a pore volume of 0.34 cm3 g−1 as measured by nitrogen the reactor was pressurized to 37 bar at a N2 flow of 30 NL h−1. At 37 bar,
sorption analysis. The Ru metal surface area, crystallite size and disper- H2 was flowed into the reactor at a flow rate of 30 NL h−1 with the N2 flow
sion were measured by oxygen chemisorption41 as 0.39 m2 g−1, 10.4 nm reduced to 9 NL h−1 such that the N2 gas in the reactor was replaced with
and 7.2%, respectively. The high purity silicon carbide (SiC) used to the H2–N2 gas mixture. This process took ~40 min as confirmed by no
dilute the catalyst bed was provided by Fiven. The gases used in the change in effluent gas composition as measured by GC. After this, the
reaction experiment—CO (purity, >99.997%), H2 (purity, >99.999%) H2 and N2 flow rates were reduced to 6 NL h−1 and 1.8 NL h−1, respectively,
and N2 (purity, >99.999%)—were obtained from BOC. and the CO gas was flowed into the reactor at the appropriate flow rate
to deliver the required feed ratio. In the first experiment, the CO flow
Operation of the reactor was set to 3 NL h−1 to give a feed ratio of 2. The gases H2, CO and N2 were
The reaction rig consists of a gas feed section, a reactor and a product used as the reaction feed gases. With these flow rates of feed gases and
collection section. The CO, H2 and N2 gases from the compressed gas the reactor pressure maintained at 37 bar, the reactor temperature was

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

increased to 220 °C and the reaction was started from a dry reactor. under these conditions. Using these combinations of observation time
After completing the MRI and 2D NMR measurements at feed ratio 2, and characteristic molecular diffusion coefficients at each feed ratio
the feed ratio was reduced to 1 by adjusting the feed H2 and N2 flow rates yields a root mean square displacement during the diffusion meas-
to 3 NL h–1 and 4.8 NL h−1, respectively, while the CO flow rate was main- urement of ~5 μm, significantly larger than the pore size of 28.6 nm,
tained at 3 NL h−1. After reaching the steady state at feed ratio 1, MRI for all feed ratios. Further increase of Δ did not influence the results.
and 2D NMR measurements were carried out. The feed ratio was then A recycle time of 3 s was applied in the diffusion measurements, which
changed to 0.5 by adjusting the feed H2 and N2 flow rates to 1.5 NL h–1 is comparable to the average T1 relaxation time of the reaction mixture.
and 6.3 NL h−1, respectively, with the CO flow rate maintained at 3 NL h−1. The effect of T1 relaxation weighting on the resulting diffusion coef-
After the measurement at feed ratio 0.5, the feed ratio was adjusted ficient and carbon number distributions was considered negligible as
back to 2 following the same operation procedure to check catalyst reported by Williams44. Other parameters used in the pulse sequences
stability. A constant weight hourly space velocity of 4 NL h−1 gcat−1 are summarized in Supplementary Tables 1 and 2.
(gcat, grams of catalyst) was used during the experiment. The total TOS The 2D COSY and double quantum filtered COSY (DQF-COSY)
of the reaction to complete all experiments was >3,000 h. No change data were acquired to determine the average carbon number of the
in the catalyst bed position throughout the experiment was confirmed products using the Bruker default pulse sequences ‘cosygpqf’ and
by comparing the images acquired at the beginning and the end of the ‘cosygpmfqf’, respectively. In both measurements, the direct and indi-
experiment (Supplementary Fig. 7). rect 1H frequency dimensions were sampled using 1,024 × 256 complex
points, respectively. Other parameters are listed in Supplementary
GC measurement Table 1. The COSY and DQF-COSY data were not acquired with spatial
GC was used to characterize the carbon number distributions of the wax resolution because of signal-to-noise ratio limitations.
products collected from the reactor exit and the composition of the The measurement of T1 relaxation time was achieved using a satu-
reactor effluent gas mixture. For the characterization of wax composi- ration recovery sequence45. A list of variable delays increasing expo-
tions, a Thermo Scientific (Trace GC-Ultra) GC instrument equipped nentially from 1 ms to 40 s in 64 steps was used for T1 encoding. Other
with a flame ionization detector was used. The effluent gas composi- parameters are reported in Supplementary Table 1. The T1 measurement
tions were measured using an Agilent Refinery Gas Analyser (Fast RGA was performed without spatial resolution.
7890B-0378), which contains three detection channels (two thermal With respect to relaxation contrast, the spin–spin relaxation
conductivity detectors and one flame ionization detector). The GC times, T2 and T2*, of the intra-pellet products are 100–300 ms and
was calibrated to measure compositions to an accuracy of ±0.5 mol%. ~3 ms, respectively. The relaxation time of liquid in the inter-pellet
space, which is in contact with SiC particles, is at least one order of
Magnetic resonance measurements magnitude faster than the intra-pellet values for both relaxation times.
The NMR measurements were performed on a Bruker Avance III HD Relevant echo times of the pulse sequences used are given in Supple-
spectrometer equipped with a vertical super-wide-bore 7.1 T super- mentary Table 1. Only T2 weighting is relevant to these acquisitions.
conducting magnet and a three-axis gradient set providing a maximum Given the T2 distribution of intra-pellet products and the echo times of
gradient strength of 83.24 G cm−1 in each direction. A birdcage r.f. coil the diffusion sequences, the bias to the diffusion and carbon number
of 66 mm inner diameter was used and tuned to a resonance frequency distributions caused by the range of T2 weighting for different species
of 300.14 MHz for 1H observation. The 1H NMR/MRI data were acquired is estimated to be no greater than one carbon number. In the presence
with a typical 1H 90° pulse duration of 65 μs. of significant field inhomogeneity, the signal of the gas-phase species in
The total signal intensity of a catalyst layer (Fig. 3b) was measured the inter-pellet space is negligible due to low 1H density and fast
using a 1D imaging sequence (default Bruker ‘improf’ sequence). The signal relaxation46.
2D slice-selective intensity images of the three catalyst layers were
acquired using a rapid acquisition with relaxation enhancement (RARE) Data processing for the magnetic resonance measurements
sequence42 with soft pulses applied at the beginning and the end of the The 1D, 2D and 3D intensity images were processed by 1D, 2D and 3D
sequence for slice selection, as shown in Supplementary Fig. 8a. The Fourier transform, respectively. For the images acquired with APGSTE
Gaussian-shape slice-selective soft pulses were 512 μs in duration and preconditioning, a diffusion-weighted 1D or 2D image was obtained at
were applied to select horizontal slices with a slice thickness of 3 mm. each value of gradient strength, g, by 1D or 2D Fourier transform. A
The positions of the selected slices for the three catalyst layers are indi- pulsed field gradient decay was obtained at each image pixel, and an
cated in Supplementary Fig. 9. The 3D image of the reactor was acquired inverse Laplace transform was applied to these data to yield a diffusion
using the Bruker default RARE sequence. The parameters used in these distribution from which the average diffusivity, D , was calculated. The
pulse sequences are summarized in Supplementary Tables 1 and 2. inverse Laplace transform was performed using the method of Tik-
The diffusion coefficient, D, values of catalyst layer 1 during the honov regularization47. In the implementation of Tikhonov regulariza-
reaction start-up (Fig. 3c) were measured using the 13-interval alternat- tion, a smoothing parameter must first be determined. In this work,
ing pulsed gradient stimulated echo (APGSTE) sequence43 followed by this parameter was determined using the method by Butler–Reeds–
1D imaging along the reactor axial direction (Supplementary Fig. 8b). Dawson, which is robust for data with a limiting signal-to-noise ratio47.
The 2D images of diffusion coefficients were acquired with a 2D RARE For the results of COSY and DQF-COSY, sine window functions were
sequence preconditioned with the APGSTE sequence (Supplementary applied to both time-domain dimensions, and the indirect dimension
Fig. 8c). The 2D images were acquired from 3 mm slices along the axial was then zero filled to 1,024 points. The resulting 2D dataset was Fou-
direction of the reactor (Supplementary Fig. 9). All the diffusion meas- rier transformed to obtain the 2D spectrum, which was then modulus
urements were performed with a r.f. pulse spacing time, τ, of 4.3 ms. corrected and symmetrized against the diagonal of the spectrum.
The pulsed gradient was applied along the x direction, perpendicular Baseline correction was then applied. The analysis of the 2D spectra
to the axial direction of the reactor, with a gradient stabilization time of was performed on the 1D anti-diagonal spectra, which were obtained
1 ms. The parameters for pulsed field gradient measurements, includ- by integrating the 2D spectra along the diagonals over the whole chemi-
ing pulse duration, δ, diffusion observation time, Δ, and pulsed gradient cal shift range32.
strength, g, used at different feed ratios are summarized in Supplemen- For the T1 relaxation measurement, the time-domain data acquired
tary Table 3. A diffusion observation time of 20 ms was used at a feed at each delay value were Fourier transformed to produce a spectrum.
ratio of 2. The observation time was increased to 40 ms for feed ratios For each value of chemical shift, a T1 relaxation recovery dataset was
1 and 0.5 because of the lower molecular diffusion coefficients observed then extracted and processed using the inverse Laplace transform to

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

yield a distribution of relaxation times associated with that chemical acquired from the reactor were severely distorted by an artefact along
shift value. In this way, a 2D chemical shift versus T1 relaxation spectrum the indirect dimension, commonly known as the t1 artefact. This arte-
was obtained. fact was significantly reduced in the DQF-COSY spectra obtained from
the same system. Therefore, in this work, the cross peak intensity was
Error analysis obtained from the DQF-COSY spectra while the diagonal peak intensity
The APGSTE-preconditioned RARE imaging of each catalyst layer was was still acquired from the COSY spectra. Hence, the value of parameter
repeated three times at the steady state at feed ratio 2. The repeat b from the original reference32 is used directly, whereas the value of
measurements were used to estimate the error of D and NC. In the 2D parameter a needs to be adjusted to account for the difference between
images of the catalyst layers, pixels that had poor data quality due to the signal intensities of the cross peaks of the DQF-COSY and COSY
low signal-to-noise ratio were identified as those where the standard spectra. The parameter a was determined using the DQF-COSY and
deviations of NC were larger than five carbon numbers. These pixels COSY spectra of pure n-alkanes with carbon numbers in the range 6–16
were removed from the D and NC maps. The error of NC calculated based confined in the titania support of the FT catalyst and at ambient tem-
on 2D NMR spectra was estimated by considering error propagation perature. The calibration data are shown in Supplementary Fig. 11a,
of the uncertainties of fit parameters in equation (3) and of experimen- and equation (3) was fitted to these data by setting the parameter b as
tal measurement. The standard deviations (s.d.) of water weight frac- the reported value32. The fit curve is also presented in Supplementary
tions were estimated using 29, 80 and 33 repeat measurements (n = 29, Fig. 11a. The resulting values of a = 0.362 ± 0.007 and b = 1.44 ± 0.02
80 and 33) at the steady state for a feed ratio of 2, 1 and 0.5, respectively. (uncertainties are standard errors of the fit) were used to determine
The results are reported as mean values ± s.d. the average carbon numbers present in the reacting system. The IXP and
Idiag values were obtained by integrating the anti-diagonal spectra of
Transformation of diffusion to carbon number distributions the 2D spectra. The anti-diagonal spectra of the DQF-COSY and COSY
The technique described by Williams et al.23 was used to transform spectra shown in Supplementary Fig. 11b,c are presented in Supple-
the distributions of molecular diffusion coefficient obtained from mentary Fig. 11d,e, respectively. The cross peak in Supplementary
the APGSTE-preconditioned imaging measurements to carbon num- Fig. 11d and the diagonal peak in Supplementary Fig. 11e, which were
ber distributions. A relationship between the carbon number Ni of a used to calculate IXP and Idiag, are highlighted in the figures.
component in a mixture of linear hydrocarbons can be expressed as a The signal detected in DQF-COSY and COSY measurements was
function of its diffusion coefficient Di (ref. 24): predominately associated with intra-pellet products; the signal associ-
1 ∞
ated with inter-pellet products was negligible due to the significant T2
−β
A v 1 v+β
−1
relaxation effects. This was confirmed by a 2D NMR experiment that
Ni = ( ) (A v φ) , φ = ∫ Di v P (Di ) dDi , (2)
Di selectively measured the data from the reactor regions packed with SiC
0
particles; negligible signal was detected from these regions. Therefore,
the average carbon number determined by COSY and DQF-COSY was
where A, β and the Flory exponent, v, are parameters that depend on associated with intra-pellet products.
temperature and the degree of confinement of the mixture but are
independent of pressure; P is a probability density function. Data availability
Following the methodology described by Williams et al.23, the The data that are used to produce the main text figures are available at
parameters A, β and v were calibrated at the reaction temperature https://doi.org/10.17863/CAM.92264. All data in the study are available
of 220 °C using liquid mixtures (or waxes) confined within the same from the corresponding author upon reasonable request.
porous titania as the catalyst support material used in the FT experi-
ments. Two wax mixtures, consisting of known distributions of chain References
lengths of n-alkanes, were used in the calibration. Their carbon number 1. Grunwaldt, J. D., Hannemann, S., Schroer, C. G. & Baiker,
distributions, determined by GC, are shown in Supplementary Fig. 10; A. 2D-mapping of the catalyst structure inside a catalytic
to the nearest integer, the average carbon number of each is 35 and 42, microreactor at work: partial oxidation of methane over Rh/Al2O3.
respectively. The data were used to calibrate the parameters in equa- J. Phys. Chem. B 110, 8674–8680 (2006).
tion (2); the resulting parameters are A = 2.22 × 10−7 m2 s−1, β = 0.27 and 2. de Smit, E. et al. Nanoscale chemical imaging of a working
v = 1.41. The carbon number distributions obtained based on these catalyst by scanning transmission X-ray microscopy. Nature 456,
calibrated parameters are plotted in Supplementary Fig. 10, from which 222–225 (2008).
average carbon numbers of 34 and 41 are obtained, consistent with the 3. Nijhuis, T. A., Tinnemans, S. J., Visser, T. & Weckhuysen, B.
GC analysis. The calibration experiments also allow the determination M. Towards real-time spectroscopic process control for the
of a noise threshold value of 0.086 required in the Butler–Reeds–Daw- dehydrogenation of propane over supported chromium oxide
son method, which was applied to determine the smoothing parameter catalysts. Chem. Eng. Sci. 59, 5487–5492 (2004).
employed in Tikhonov regularization47. 4. Urakawa, A., Maeda, N. & Baiker, A. Space- and time-resolved
combined DRIFT and Raman spectroscopy: monitoring dynamic
Determination of average carbon numbers using 2D NMR surface and bulk processes during NOx storage reduction. Angew.
The ratio between the intensities of the cross and diagonal peaks IXP/Idiag Chem. Int. Ed. 47, 9256–9259 (2008).
of the COSY spectra of n-alkanes or n-alkane mixtures has been 5. Stavitski, E., Kox, M. H. F., Swart, I., de Groot, F. M. F.
reported to decrease with increasing carbon number NC or average & Weckhuysen, B. M. In situ synchrotron-based IR
carbon number NC following a power law relationship independent of microspectroscopy to study catalytic reactions in zeolite crystals.
temperature32: Angew. Chem. Int. Ed. 47, 3543–3547 (2008).
6. Ulpts, J., Dreher, W., Klink, M. & Thoming, J. NMR imaging of gas
−b
IXP /Idiag (NC ) = a (NC ) . (3) phase hydrogenation in a packed bed flow reactor. Appl. Catal.
A Gen. 502, 340–349 (2015).
7. Baker, L. et al. Operando magnetic resonance studies of phase
The parameters a and b have been determined using the COSY behaviour and oligomer accumulation within catalyst pores
spectra of n-alkanes and n-alkane mixtures for which NC and NC were in during heterogeneous catalytic ethene oligomerization. Appl.
the range 5–40 (ref. 32). However, the cross peaks of the COSY spectra Catal. A Gen. 557, 125–134 (2018).

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

8. Weckhuysen, B. M. Chemical imaging of spatial heterogeneities in 28. Schurm, L., Kern, C. & Jess, A. Accumulation and distribution
catalytic solids at different length and time scales. Angew. Chem. of higher hydrocarbons in the pores of a cobalt catalyst during
Int. Ed. 48, 4910–4943 (2009). low-temperature Fischer–Tropsch fixed-bed synthesis. Catal. Sci.
9. Meirer, F. & Weckhuysen, B. M. Spatial and temporal exploration Technol. 11, 6143–6154 (2021).
of heterogeneous catalysts with synchrotron radiation. Nat. Rev. 29. Pinard, L. et al. Identification of the carbonaceous compounds
Mater. 3, 324–340 (2018). present on a deactivated cobalt based Fischer–Trospch catalyst
10. Weber, D., Holland, D. J. & Gladden, L. F. Spatially and chemically resistant to “rejuvenation treatment”. Appl. Catal. A Gen. 406,
resolved measurement of intra- and inter-particle molecular 73–80 (2011).
diffusion in a fixed-bed reactor. Appl. Catal. A Gen. 392, 30. Bukur, D. B., Mandic, M., Todic, B. & Nikacevic, N. Pore diffusion
192–198 (2011). effects on catalyst effectiveness and selectivity of cobalt based
11. Bouchard, L. S. et al. NMR imaging of catalytic hydrogenation Fischer-Tropsch catalyst. Catal. Today 343, 146–155 (2020).
in microreactors with the use of para-hydrogen. Science 319, 31. Krishnamoorthy, S., Tu, M., Ojeda, M. P., Pinna, D. & Iglesia, E. An
442–445 (2008). investigation of the effects of water on rate and selectivity for the
12. Jarenwattananon, N. N. et al. Thermal maps of gases in Fischer–Tropsch synthesis on cobalt-based catalysts. J. Catal. 211,
heterogeneous reactions. Nature 502, 537–540 (2013). 422–433 (2002).
13. Burueva, D. B. et al. In situ monitoring of heterogeneous catalytic 32. Terenzi, C., Sederman, A. J., Mantle, M. D. & Gladden, L. F.
hydrogenation via 129Xe NMR spectroscopy and proton MRI. ACS Enabling high spectral resolution of liquid mixtures in porous
Catal. 10, 1417–1422 (2020). media by antidiagonal projections of two-dimensional 1H NMR
14. Kuppers, M., Heine, C., Han, S., Stapf, S. & Blumich, B. COSY spectra. J. Phys. Chem. Lett. 10, 5781–5785 (2019).
In situ observation of diffusion and reaction dynamics in gel 33. Mandic, M., Todic, B., Zivanic, L., Nikacevic, N. & Bukur, D. B.
microreactors by chemically resolved NMR microscopy. Appl. Effects of catalyst activity, particle size and shape, and process
Magn. Reson. 22, 235–246 (2002). conditions on catalyst effectiveness and methane selectivity for
15. Nicholas, C. P. Applications of light olefin oligomerization to the Fischer–Tropsch reaction: a modeling study. Ind. Eng. Chem. Res.
production of fuels and chemicals. Appl. Catal. A Gen. 543, 56, 2733–2745 (2017).
82–97 (2017). 34. Hubble, R., York, A. P. E. & Dennis, J. S. Modelling reaction and
16. James, O. O., Chowdhury, B., Mesubi, M. A. & Maity, S. Reflections diffusion in a wax-filled hollow cylindrical pellet of Fischer
on the chemistry of the Fischer–Tropsch synthesis. RSC Adv. 2, Tropsch catalyst. Chem. Eng. Sci. 207, 958–969 (2019).
7347–7366 (2012). 35. Marano, J. J. & Holder, G. D. Characterization of Fischer-Tropsch
17. Wood, D. A., Nwaoha, C. & Towler, B. F. Gas-to-liquids (GTL): liquids for vapor-liquid equilibria calculations. Fluid Phase Equilib.
a review of an industry offering several routes for monetizing 138, 1–21 (1997).
natural gas. J. Nat. Gas. Sci. Eng. 9, 196–208 (2012). 36. Pohlmann, F., Kern, C., Rossler, S. & Jess, A. Accumulation of
18. Ail, S. S. & Dasappa, S. Biomass to liquid transportation fuel liquid hydrocarbons in catalyst pores during cobalt-catalyzed
via Fischer Tropsch synthesis – technology review and current Fischer–Tropsch synthesis. Catal. Sci. Technol. 6, 6593–6604
scenario. Renew. Sust. Energ. Rev. 58, 267–286 (2016). (2016).
19. Gao, R. X. et al. Green liquid fuel and synthetic natural gas 37. Bezemer, G. L., Remans, T. J., van Bavel, A. P. & Dugulan, A. I.
production via CO2 hydrogenation combined with reverse Direct evidence of water-assisted sintering of cobalt on carbon
water-gas-shift and Co-based Fischer-Tropsch synthesis. J. nanofiber catalysts during simulated Fischer–Tropsch conditions
CO2 Util. 51, 101619 (2021). revealed with in situ Mössbauer spectroscopy. J. Am. Chem. Soc.
20. Iglesia, E. Design, synthesis, and use of cobalt-based 132, 8540–8541 (2010).
Fischer-Tropsch synthesis catalysts. Appl. Catal. A Gen. 161, 38. Kliewer, C. E., Soled, S. L. & Kiss, G. Morphological
59–78 (1997). transformations during Fischer-Tropsch synthesis on a
21. Pohlmann, F. & Jess, A. Interplay of reaction and pore diffusion titania-supported cobalt catalyst. Catal. Today 323,
during cobalt-catalyzed Fischer–Tropsch synthesis with CO2-rich 233–256 (2019).
syngas. Catal. Today 275, 172–182 (2016). 39. Zheng, Q. et al. Water-wax behaviour in porous silica at low
22. Papavasileiou, K. D. et al. Molecular dynamics simulation of temperature Fischer–Tropsch conditions. Appl. Catal. A Gen. 572,
the n-octacosane-water mixture confined in hydrophilic and 142–150 (2019).
hydrophobic mesopores: the effect of oxygenates. Fluid Phase 40. Liu, G., Li, Y. & Jonas, J. Confined geometry effects on
Equilib. 526, 112816 (2020). reorientational dynamics of molecular liquids in porous silica
23. Williams, J. et al. In situ determination of carbon number glasses. J. Chem. Phys. 95, 6892–6901 (1991).
distributions of mixtures of linear hydrocarbons confined within 41. Taylor, K. C. Determination of ruthenium surface areas by
porous media using pulsed field gradient NMR. Anal. Chem. 92, hydrogen and oxygen chemisorption. J. Catal. 38,
5125–5133 (2020). 299–306 (1975).
24. Freed, D. E., Burcaw, L. & Song, Y. Q. Scaling laws for diffusion 42. Hennig, J., Nauerth, A. & Friedburg, H. RARE imaging: a fast
coefficients in mixtures of alkanes. Phys. Rev. Lett. 94, 067602 imaging method for clinical MR. Magn. Reson. Med. 3,
(2005). 823–833 (1986).
25. Madon, R. J. & Iglesia, E. The importance of olefin readsorption 43. Cotts, R. M., Hoch, M. J. R., Sun, T. & Markert, J. T. Pulsed field
and H2/CO reactant ratio for hydrocarbon chain growth on gradient stimulated echo methods for improved NMR diffusion
ruthenium catalysts. J. Catal. 139, 576–590 (1993). measurements in heterogeneous systems. J. Magn. Reson. 83,
26. Kuipers, E. W., Vinkenburg, I. H. & Oosterbeek, H. Chain-length 252–266 (1989).
dependence of α-olefin readsorption in Fischer-Tropsch 44. Williams, J. Magnetic Resonance Studies of the Diffusion Dynamics
synthesis. J. Catal. 152, 137–146 (1995). of Molecular Systems Relevant to Fischer-Tropsch Catalysis. PhD
27. Lillebo, A., Rytter, E., Blekkan, E. A. & Holmen, A. Fischer–Tropsch thesis, Univ. of Cambridge (2020).
synthesis at high conversions on Al2O3-supported Co catalysts 45. Markley, J. L., Horsley, W. J. & Klein, M. P. Spin-lattice relaxation
with different H2/CO levels. Ind. Eng. Chem. Res. 56, 13282–13287 measurements in slowly relaxing complex spectra. J. Chem. Phys.
(2017). 55, 3604–3605 (1971).

Nature Catalysis
Article https://doi.org/10.1038/s41929-023-00913-8

46. Mitchell, J., Chandrasekera, T. C., Johns, M. L., Gladden, L. F. J.W., T.A.B., G.L.B., C.M.G., M.D.M., A.J.S. and L.F.G. edited and revised
& Fordham, E. J. Nuclear magnetic resonance relaxation and the manuscript.
diffusion in the presence of internal gradients: the effect of
magnetic field strength. Phys. Rev. E 81, 026101 (2010). Competing interests
47. Mitchell, J., Chandrasekera, T. C. & Gladden, L. F. Numerical The authors declare no competing interests.
estimation of relaxation and diffusion distributions in two
dimensions. Prog. Nucl. Magn. Reson. Spectrosc. 62, 34–50 Additional information
(2012). Supplementary information The online version contains
48. Zheng, Q. et al. Experimental determination of H2 and CO supplementary material available at
diffusion coefficients in a wax mixture confined in a porous titania https://doi.org/10.1038/s41929-023-00913-8.
catalyst support. J. Phys. Chem. B 124, 10971–10982 (2020).
Correspondence and requests for materials should be addressed to
Acknowledgements Lynn F. Gladden.
We thank P. Munnik for catalyst synthesis and J. Chen for calculations
and simulation studies supporting this work. This work was funded Peer review information Nature Catalysis thanks the anonymous
by Shell Global Solutions International B.V. Q.Z. thanks the IChemE reviewers for their contribution to the peer review of this work.
Andrew Fellowship for additional financial support. For the purpose
of open access, the authors have applied a Creative Commons Reprints and permissions information is available at
Attribution (CC BY) licence to any Author Accepted Manuscript version www.nature.com/reprints.
arising from this submission.
Publisher’s note Springer Nature remains neutral with regard to
Author contributions jurisdictional claims in published maps and institutional affiliations.
Q.Z. led the experimental design, the data acquisition and
interpretation and the writing of the manuscript. J.W. was also involved Springer Nature or its licensor (e.g. a society or other partner) holds
in the data collection, and developed and applied the analysis for exclusive rights to this article under a publishing agreement with
transforming the diffusion measurements into hydrocarbon chain the author(s) or other rightsholder(s); author self-archiving of the
length data. L.R.v.T. and S.V.E. contributed to the experimental design accepted manuscript version of this article is solely governed by the
and data acquisition, and design of the reaction engineering set-up. terms of such publishing agreement and applicable law.
T.A.B. and C.M.G. helped design the experiments and provided the
catalyst materials. L.F.G., M.D.M. and A.J.S. oversaw the project and © The Author(s), under exclusive licence to Springer Nature Limited
worked with the coauthors to analyse and interpret the data obtained. 2023

Nature Catalysis

You might also like