Characterization and catalytic behavior of CO/SiO2 catalysts: Influence of dispersion in the Fischer-Tropsch reaction

Article abstract of DOI:10.1006/jcat.2001.3204

The study of hydrogen evolution from temperature-programmed desorption (TPD) experiments appears to be so complex that it casts some doubt on the reliability of chemisorption techniques for the estimation of cobalt dispersion in a series of Co/SiO₂ catalysts. Co particle sizes were examined by magnetic measurements (MMs), transmission electron microscopy (TEM), and extended X-ray absorption fine structure (EXAFS). EXAFS shows that Co in the catalysts crystallizes in fcc crystal lattice and that relatively small clusters are formed. The deduced average size is almost an order of magnitude smaller than that obtained by MMs and TEM, suggesting a complex morphology of Co particles. Study of size sensitivity in the Fischer-Tropsch reaction shows that intrinsic activity and chain growth probability first increase and then stabilize on increasing the particle size, thus providing a quantitative determination of the critical diameter separating these two zones, i.e., ca. 6 nm. Study of the surface species after reaction by TPH experiments indicates that (i) deactivation may be due to a loss of catalytically active cobalt; (ii) in the course of the reaction, neither formates nor acetates are detected in the catalyst; and (iii) four carbon-containing species are observed (α, β, γ, and δ). Species β has been speculated to be the active phase of the reaction on the basis of its parallel variation with intrinsic activity. This intermediate, which is strongly interacting with the cobalt phase, does not contain oxygen atoms. The δ species might be associated with graphitic residue formation, whereas no specific role has been attributed to the γ species.

Full text of DOI:10.1006/jcat.2001.3204

Journal of Catalysis 200, 106–116 (2001)
doi:10.1006/jcat.2001.3204, available online at http://www.idealibrary.com on
Characterization and Catalytic Behavior of Co/SiO2 Catalysts: Influence
of Dispersion in the Fischer–Tropsch Reaction
,
,
Arnaud Barbier, ,1 Alain Tuel, ,2 Iztok Arcon,† ‡ Aloz Kodre,‡ § and Guy Antonin Martin
Institut de Recherches sur la Catalyse, CNRS, Villeurbanne, France; †Nova Gorica Polytechnic, Vipavska 13, 5000 Nova Gorica, Slovenia;
‡Institute Jozef Stefan, Jamova 39, 1000 Ljubljana, Slovenia; and §Faculty of Mathematics and Physics, University of Ljubljana,
Jadranska 19, 1000 Ljubljana, Slovenia
Received October 2, 2000; revised January 22, 2001; accepted January 22, 2001; published online April 18, 2001
fective in this reaction. The sensitivity of this reaction to Co
particle size, however, is not well understood. For Co/SiO2
catalysts, the absence of any significant particle size effect
in the range 6–20% dispersion observed by Ho et al. (1)
at atmospheric pressure contrasts with a former study by
Reuel and Bartholomew (2), who reported an increase in
the turnover frequency by a factor of 3 on decreasing disper-
sion from 20 to 10% under similar conditions and using the
same type of catalyst. The invariance of the intrinsic activity
has been confirmed by Iglesia et al. (3) over the range 3.2–
9.5% at 2000 kPa. Literature data show that chain growth
probabilityincreaseswith cobalt particle size (4, 5), whereas
this parameter is invariant for large particle sizes (3). More-
over, the influence of cobalt particle size on the stability of
the catalyst is not documented.
The uncertainty in the size sensitivity of a reaction may
be related to the limits or to the ambiguity of the method
used for determining particle size. Reuel and Bartholomew
(6) have recommended hydrogen adsorption as the most
convenient reliable technique to measure cobalt particle
size. This conclusion was based on a comparison between
chemisorption data and transmission electron microscopy
results or BET surface area measurements performed on a
series of supported and unsupported catalysts, respectively.
This technique is now widely used to characterize cobalt
dispersion. We have previously reported (7) that the sur-
face average diameters calculated by magnetic techniques
are in good agreement with those obtained by transmission
electron microscopy for a series of Co/SiO2 catalysts pre-
pared by the ammonia method and having average sizes
in the range ca. 4–10 nm. Extended X-ray absorption fine
structure (EXAFS) spectroscopy, which provides structural
information on the local order around a selected type of
atom (8), can be used to characterize the arrangement of
metal atoms in dispersed metal catalysts. Crystal structure
of metallic clusters and their average size can be deduced
from EXAFS data (9–11). However, EXAFS is not a direct
method for particle size determination, the average size be-
ing obtained from the average number of neighbors, with
The studyofhydrogen evolution from temperature-programmed
desorption (TPD) experiments appears to be so complex that it casts
some doubt on the reliability of chemisorption techniques for the
estimation of cobalt dispersion in a series of Co/SiO2 catalysts. Co
particle sizes were examined by magnetic measurements (MMs),
transmission electron microscopy (TEM), and extended X-ray ab-
sorption fine structure (EXAFS). EXAFS shows that Co in the
catalysts crystallizes in fcc crystal lattice and that relatively small
clusters are formed. The deduced average size is almost an order of
magnitudesmaller than that obtained byMMsand TEM, suggesting
a complex morphology of Co particles. Study of size sensitivity in
the Fischer–Tropsch reaction shows that intrinsic activity and chain
growth probability first increase and then stabilize on increasing the
particle size, thus providing a quantitative determination of the crit-
ical diameter separating these two zones, i.e., ca. 6 nm. Study of the
surface species after reaction by TPH experiments indicates that
(i) deactivation may be due to a loss of catalytically active cobalt;
(ii) in the course of the reaction, neither formates nor acetates are
detected in the catalyst; and (iii) four carbon-containing species are
observed ( ,
, , and ). Species has been speculated to be the ac-
tive phase of the reaction on the basis of its parallel variation with
intrinsic activity. This intermediate, which is strongly interacting
with the cobalt phase, does not contain oxygen atoms. The species
might be associated with graphitic residue formation, whereas no
specific role has been attributed to the species.
c 2001 Academic Press
INTRODUCTION
The Fischer–Tropsch synthesis whereby liquid aliphatic
hydrocarbons and oxygenates are produced by the hydro-
genation of carbon monoxide has been known for about
75 years. Cobalt-containing catalysts are known to be ef-
1 Present address: IUT of Rouen, LASTSM, Mont Saint-Aignan,
France.
2 To whom correspondence should be addressed at Institut de
Recherches sur la Catalyse, CNRS, 2, avenue A. Einstein, 69626
Villeurbanne Cedex, France. Fax: (+33) (0)4 72 44 53 99. E-mail: tuel@
catalyse.univ-lyon1.fr.
106
0021-9517/01 $35.00
c
Copyright
2001 by Academic Press
All rights of reproduction in any form reserved.

Co/SiO2 CATALYSTS IN FISCHER–TROPSCH REACTION
107
TABLE 1
an assumption on the particle shape. Corrections for partial
disorder at the surface of clusters have been used (12, 13) to
reconcile the size estimates with results of other methods.
The aim of this paper is to examine a series of well-
characterized Co/SiO2 catalysts prepared according to the
ammonia method as described in Ref. (7) to gain a better
understanding of the size sensitivity in the Fischer–Tropsch
reaction with respect to activity, selectivity, and stability.
For this purpose, temperature-programmed hydrogenation
(TPH) of species present in the catalysts and magnetic mea-
surements were carried out to better define the catalytic sys-
tems under reaction conditions. In addition, EXAFS mea-
surements were performed to gain further insight into the
intimate morphology of cobalt particles. Data obtained by
Cobalt Loading and Surface Average Diameters as Obtained
from Magnetic Measurements (MMs) and Transmission Electron
Microscopy (TEM) for Samples 1–5a
Ds (nm)
Sample
Method
Co (wt% )
MM
TEM
9.2
1b
2
3
4
5
Ammonia
Ammonia
Ammonia
Ammonia
Fluoride
3.2
8.4
8.2
19.2
1.83
3.8
5.5
6.2
5.0
a Reduction was carried out in flowing hydrogen (6 L h 1) by slowly
1
the different techniques are discussed and a model is pro- increasing the temperature at a heating rate of 1.3 K min for 8 h. The
posed to explain the apparent discrepancy observed be-
tween data obtained by EXAFS and other techniques.
final standard temperature was 923 K.
b For catalytic experiments a new batch has led to the following char-
acteristics: cobalt loading, 3.5 wt% ; surface average diameter, 4.4 nm.
CATALYST PREPARATION AND CHARACTERIZATION
Sample 5 was prepared by reacting a fluoride precur-
Samples 1 to 4 were prepared using the ammonia method sor with silica. This original method is an extension of
asdescribed elsewhere (7). Typically, a solution ofcobalt(II) that proposed to obtain well-dispersed Ti/SiO2 catalysts
nitrate protected from air by an argon blanket was first (14). Cobalt(III) fluoride (0.193 g) was added to a sol-
prepared. After stirring, a 22 wt% solution of ammonia vent (diglyme, 2-methoxyethyl ether) and the mixture was
was added dropwise followed by silica (Aerosil 200 from heated at 333 K for 30 min in a glass flask equipped with a
Degussa, 200 m2/g). After pH equilibrium, the system was cooling column. After dissolution of the cobalt fluoride, 5 g
centifuged, washed several times, and finally dried in a vac- of silica (Aerosil 800 from Degussa, 800 m2/g) was added
uum oven at 353 K overnight. The precursor was reduced and stirring was maintained at 333 K for 2 h. The solid
in flowing hydrogen (60 L h 1 g 1) by slowly increasing the phase was filtered and dried in a vacuum oven at 393 K
temperature at a heating rate of 1.3 K/min for 8 h. Samples overnight. The precursor was reduced under conditions
were held at the final temperature (923 K) for 2 h. Magnetic similar to those used for samples 1–4 except that the final re-
experiments aimed at characterizing the reduced Co/SiO2 duction temperature was 1023 K. Magnetic measurements
catalysts thus obtained were performed in situ in an elec- have shown that the precursor of sample 5 is less reducible
tromagnet with fields up to 21 kOe (2.1 T) at 300 K using than those of samples 1–4: in the former, the extents of re-
the Weiss extraction method. These experiments give in- duction are 0, 80, 82, 85, and 87% at 473, 560, 703, 910, and
teresting information on cobalt characteristics, as described 1023 K, respectively. When the reduction is carried out at
elsewhere (7). The amount of reduced cobalt in the metallic 1023 K, the cobalt particle size Ds is found to be 5 nm, as
state isobtained bycomparingthe saturation magnetization measured by magnetic measurements (Table 1). In contrast
of the sample with that of bulk Co. The corresponding ra- to samples 1–4, sample 5 is not reoxidized when contacted
tio, F, yields the extent of reduction. It has been shown that with air at 300 K. Furthermore, this solid does not show
under our reduction conditions, the precursors obtained by any catalytic activity, suggesting that cobalt particles are
the ammonia method are completely reduced (7).
embedded in silica.
Furthermore, when superparamagnetism occurs (no re-
EXAFS spectra at the Co K edge were measured at
manence after saturation) average diameters of cobalt par- EXAFS II station in HASYLAB at DESY. The station
ticles can be calculated from the variations in magnetization provides a focused beam from an Au-coated mirror and
with field strength at low and high fields (D1 and D2, respec- a Si(111) double-crystal monochromator with 1.5-eV reso-
tively). The surface average diameter, Ds, is then defined as lution at the Co K edge. Harmonics are effectively elimi-
Ds = (D1 + D2)/2. It has been shown that surface average nated by a plane Au-coated mirror and by a slight detun-
diameters thus obtained are in good agreement with those ing of the monochromator crystals, keeping the intensity
calculated from transmission electron microscopy (7). The at 60% of the rocking curve with the beam stabilization
Ds values for samples 1–3 are reported in Table 1. The sur- feedback control. Due to the low concentration of cobalt
face average diameter for sample 4, which is not super- in the samples, about 1 mm of powder was compressed into
paramagnetic, has been obtained by electron microscopy a liquid absorption cell with Kapton windows to obtain a
(Table 1).
homogeneous layer with a total absorption thickness of 1.5

108
BARBIER ET AL.
above the Co K edge, with a Co K-edge jump of ca. 0.5. The
empty absorption cell served as the reference spectrum un-
der identical conditions. The EXAFS spectrum of Co metal
foil was also measured for comparison.
CATALYSIS AND TEMPERATURE-PROGRAMMED
HYDROGENATION
The Fischer–Tropsch reaction was carried out in a fixed-
bed reactor at atmospheric pressure using 1 g of catalyst
reduced in situ under standard conditions (conversion less
than a few percent). Gases were analyzed on-line by gas
chromatography (FID and TCD) at the reactor outlet. The
feed ratio CO/H2 was 0.5. The CO gas flow per cobalt weight
was 0.1 mL min 1 g 1. The reaction temperatures were suc-
cessively 443, 453, 463, and 473 K; the catalyst was held at
each temperature for 2 h. The deactivation was estimated
from the loss of activity at 453 K for 24 h. After reaction, the
catalyst was cooled down at room temperature in a helium
flow, then heated to 1023 K at a heating rate of 20 K/min in
a mixture of hydrogen (3% ) and argon with a flow rate of
60 mL/min. The gas composition at the outlet of the reactor
was continuously analyzed in a VG-quadrupole mass spec-
trometer. The most abundant reaction products were H2O,
CH4, and CO. Small amounts of CO2, C2H6, and C3H8 were
also detected, whereas O2 and CH3OH were not detected.
A series of temperature-programmed desorptions (TPDs)
of hydrogen were also performed on reduced samples be-
fore reaction. The experimental conditions were similar to
those for TPH, except that the argon flow was free of hy-
drogen.
FIG. 1. Profiles of hydrogen temperature-programmed desorption
for samples 1, 3, and 4: normalized intensity as a function of the TPD
temperature.
spectra of samples 1, 4, and 5 are shown in Fig. 3. Fourier
transforms of the EXAFS spectra (Fig. 4) show the typical
local structure of metallic Co. Comparison with the spec-
trum measured on Co metal foil reveals the same pattern of
neighbors. However, the size of individual neighbor peaks
is lower than in the Co foil, suggesting that the average
number of neighbors is decreased and hence that clusters
of Co metal are formed. From the observed reduction of
the average number of neighbors, the size of the clusters
can be deduced (9–11). For sample 1, an estimate of the
size can also be given by the cutoff in the FT spectrum: the
˚
absence of a prominent peak observed at 7.5 A in the FT
spectrum of bulk Co provides an upper limit.
RESULTS
Cobalt Dispersion from Hydrogen TPD
The hydrogen temperature desorptions performed on
samples 1, 3, and 4 show three sets of peaks at low, medium,
and high temperature (l, m, and h, respectively) (Fig. 1).
The m peak is observed only for sample 1. The atomic ratio
Nx = H/Cos (where H is the number of hydrogen atoms,
Cos the number of surface cobalt atoms as calculated from
Ds, assuming a cubo-octahedral shape, and x the hydrogen
peak nature) depends on the size of cobalt particles and is
found to be smaller than unity (Fig. 2). Nl goes through a
maximum for Ds = 6.2 nm while Nh decreases linearly when
Ds increases. N(l+m+h) decreases by a factor of 5 when Ds
increases from 4 to 9 nm. These results show that the eval-
uation of cobalt particle size by hydrogen chemisorption,
frequently used in the literature, leads to unreliable results.
EXAFS
FIG. 2. Atomicratio Nx = Hx/Cos (where H isthe number ofhydrogen
atoms, Cos the number of surface cobalt atoms as calculated from magnetic
measurements, and x refers to the hydrogen peak nature, l (᭡), h (᭹), or
The EXAFS spectra were analyzed by the UWXAFS
code (15, 16) in the K range 5–12 A 1, using k3 weight and
˚
a Hanning window. The k3-weighted Co K-edge EXAFS l + m + h (+)) as a function of the surface average diameter (Ds).

Co/SiO2 CATALYSTS IN FISCHER–TROPSCH REACTION
109
FIG. 3. k3-weighted Co K-edge EXAFS spectra of samples 1, 4, and
5. Dots, experiments; solid line, fit.
FIG. 4. k3-weighted Fourier transform magnitude of Co K-edge EX-
AFS spectra of Co metal foil and samples 1, 4, and 5. Solid line, experimen-
tal; dashed line, fcc model. With the spectrum of Co foil, the hcp model is
also shown (upper spectrum).
To describe the clusters, FEFF models(15) ofboth known
Co metal lattices, fcc and hcp, are constructed. The two lat-
tices differ only in the stacking order of dense (fcc 111)
planes, so that the immediate neighborhood of an atom is
the same in both, and the differences arise only in parame-
ters of the second and further neighbors. The fcc model with ment with experimental spectra in the first-neighbor peak
˚
the lattice constant a = 3.61 A comprises the four short- is perfect but further peaks show a significant discrepancy
est single scattering paths of the lattice and all multiple (Fig. 4). We take this fact as proof of the fcc lattice in the
˚
scattering paths up to 5.8 A. The same range of scattering calibrating Co foil and, consequently, in the samples.
paths is used in the hcp model (a = 2.51 A, c = 4.07 A). The
In modeling the clusters, the fixed value of S02 (0.86(3))
˚
˚
2
1
Co foil spectrum is used to calibrate the models. For fcc, is retained, while DT,
and 1 r/r are varied, and the fcc
an excellent fit is obtained in the region between 1.5 and shell coordination numbers in the model are multiplied by
˚
5.0 A with only four variable parameters: the lattice ex- variable average neighbor fractions(ni). The accuracyofthe
pansion (1 r/r), the Debye temperature (DT) in modeling results for the variable parameters of the model (Table 2)
the Debye–Waller factors ( 2) of all paths (17), a separate is supported by the good agreement with data (Fig. 4). For
2
for the first-neighbor shell, and the amplitude reduc- sample 1, an additional shell containing on average 0.8(3)
1
2
2
tion factor (S02). The shell coordination numbers and radius oxygen atom at r = 1.97(3) A with
= 0.004(1) A has to
˚
˚
ratios are fixed at their fcc values. For the hcp model, agree- be introduced for a fit of equivalent quality.
TABLE 2
Average Neighbor Fractions (ni) of Co Neighbors in the First Four Coordination Shells in the Co/SiO2 Samples a
2
2
1
˚
(A )
TD (K)
Sample
n1
n2
n3
n4
1 r/r
shell 1
shells 2–4
1
4
5
0.52(4)
0.66(5)
0.72(4)
0.37(8)
0.53(9)
0.63(6)
0.21(5)
0.42(6)
0.65(4)
0.15(5)
0.38(8)
0.64(4)
0.037(4)
0.048(4)
0.053(4)
0.010(1)
0.0085(6)
0.0074(5)
300(90)
350(20)
340(20)
N1
N2
N3
N4
Co metal
12
6
24
12
0.047(4)
0.0068(2)
364(10)
a Best-fit values for the lattice expansion (1 r/r), the Debye–Waller factor of the first coordination shell 12, and the Debye
temperature (TD) governing the width of further shells are added. The model parameters determined on the reference Co
metal foil are added for comparison. Shell coordination numbers Ni were fixed at their fcc values. Uncertainties on the last
digit are given in parentheses.

110
BARBIER ET AL.
TABLE 3
do not show significant contraction of the interatomic dis-
tances compared with bulk Co.
Model: Average Number Fraction (ni) of Consecutive Neighbors
in fcc Magic-Number Co Clusters
Catalysis
Cluster
Nat/cl
diameter (nm)
n1
n2
n3
Figure 5 shows the variation in intrinsic activity (turnover
frequency (TOF), number of CO molecules converted per
surface cobalt atom, as deduced from Ds assuming a cubo-
octahedral shape, per second, measured at 453 K) and of
the specific activity (number of CO molecules converted
per total cobalt atom per second at 453 K) with the sur-
face average diameter of cobalt particles as obtained by
magnetic measurements after reaction. The average diam-
eter does not exactly coincide with that reported in Table 1,
measured before reaction, probably because some sinter-
ing has occurred during reaction. As can be seen, the TOF
increases continuously with Ds, first rapidly in the region
4–6 nm, then more slowly from 6 to 9 nm. Chain growth
probability, , is calculated from the slope of the curve
ln(Sn/n) versus n, where n is the carbon number of the hy-
drocarbon and Sn the corresponding selectivity. As shown
in Fig. 6, increases when the temperature decreases. Fur-
thermore, chain growth probabilityisan increasingfunction
of cobalt particle diameter: this parameter first increases
rapidly and then more slowly for diameters larger than
6 nm.
13
19
43
0.511
0.722
0.884
1.021
1.351
1.614
1.976
2.0
0.4615
0.5263
0.6047
0.6545
0.7407
0.7861
0.818
0.808
0.870
0.904
0.3077
0.3158
0.5116
0.5455
0.6667
0.6567
0.737
0.729
0.819
0.864
0.1538
0.2105
0.3721
0.4000
0.5481
0.6070
0.678
0.668
0.778
0.834
55
135
201
344
380a
1300a
3000a
3.0
4.0
a Analytical estimation of N
(from (10)).
and n for a given cluster diameter
at /cl
The average neighbor fractions (ni) of the first four shells
extracted from EXAFS spectra of the samples are con-
verted into cluster sizes with the following assumption.
Model clusters are constructed as fcc magic-number clus-
ters by filling consecutive crystallographic shells, i.e., shells
of sites with all permutations of the same crystallographic
indices with respect to the central atom. The model neigh-
bor fractions are obtained by summing the number of ith
neighbors over all atoms of the cluster and dividing by the
corresponding number in the bulk (Table 3). The experi-
mental neighbor fractions are compared with the table of
values from model clusters of increasing size (Table 3). The
best agreement is obtained for clusters containing 19, 55,
and 135 atoms (diameters of 0.72, 1.02, and 1.35 nm) for
samples 1, 4, and 5, respectively. The agreement in all four
shells for the two smaller clusters is remarkable. It is possi-
ble that the larger deviation for sample 5 indicates a defor-
mation from the globular shape.
The influence of cobalt particle size on the rate of de-
activation is illustrated in Fig. 7. As Ds increases, this rate
first increases, goes through a maximum for D = 5.5 nm,
and then decreases. Over the range studied, catalysts with
a particle size of ca. 10 nm seem to be the most stable.
From the sizes obtained it becomes clear why the oxygen
neighbors are found only for sample 1. They probably come
from the contact layer with the substrate and become visible
only for the smallest clusters where most of the Co atoms
in the cluster are located on its surface.
Further support for the cluster sizes obtained based on
the spherical cluster model is provided by the analysis of
Debye–Waller factors and Co–Co distances. An increase in
the Debye–Waller factor for the first shelland a correspond-
ing decrease in the DT for more distant shells compared
with bulk Co are observed for all samples. The increase in
the Debye–Waller factor is higher for smaller clusters and
can be ascribed to the increased static part of the disorder
in the clusters (11, 17). A smaller reduction of interatomic
distances is observed only in the case of small clusters (sam-
ple 1), in agreement with similar observations on small Cu
FIG. 5. Variation of the intrinsic activity ((᭡), turnover frequency,
TOF, number of CO molecules converted per surface cobalt atoms per
second measured at 453 K) and the specific activity ((᭹), number of CO
molecules converted per total cobalt atom per second at 453 K) with the
clusters (17, 18). Larger clusters found in samples 4 and 5 surface average diameter (Ds).

Co/SiO2 CATALYSTS IN FISCHER–TROPSCH REACTION
111
FIG. 6. Variation of chain growth probability ( ) at different temper-
atures (443 K (+), 453 K (᭡) and 473 K (᭹)) with the surface average
diameter (Ds).
TPH Analysis of Species Present in Sample 4
after 24 h of Reaction
Figure 8 shows the product evolution of the TPH temper-
ature for sample 4 after 24 h of reaction and Table 4 reports
the order of magnitude of the mass spectrometer intensity
and the maximum temperature of the signal, Tm. As can
be seen, the hydrogen consumption peak (435 K) and the
first water evolution peak (440 K) are observed to coin-
cide. Furthermore, the corresponding peak intensities are
comparable. This observation strongly suggests that these
peaks result from the hydrogenation of oxygen atoms is-
sued from the carbon monoxide dissociation. Thisreduction
takes place at a temperature lower than that corresponding
to hydrocarbon evolution (435–440 K vs 451–461 K), sug-
FIG. 8. TPH profiles of sample 4 after 24 h of reaction: intensity in
arbitrary units (a.u.) as a function of the TPH temperature.
gesting that oxygen atom removal is not the rate-limiting
step in the Fischer–Tropsch reaction.
The origin of hydrogen peak formation at 640 K, which
does not correspond to any hydrogen peak desorption in
TABLE 4
Order of Magnitude of the Mass Spectrometer Intensity and
Maximum Temperature of the Signal, Tm
Intensity order
of magnitude
Tm
(K)
Products
8
Consumption
Formation
Formation
Formation
Formation
Formation
Formation
Formation
Formation
Formation
H2
H2
7
7
3
10
10
10
435
660
8
8
H2O
CH4
CO
CO2
C2H6
C3H8
O2
440, 550–650
461, 644, 827
500, 644, 950
422, 510, 664
456, 628
451, 619
—
8
1.5 10
7
9
10
9
1.5 10
1.5 10
1
1
1
9
9
10
10
10
12
12
FIG. 7. Deactivation (estimated from the loss of activity at 453 K for
24 h) as a function of surface average diameter (Ds).
CH3OH
—

112
BARBIER ET AL.
480 K, respectively. A broad peak of water is observed at
low temperature, indicative of its accumulation in the cata-
lyst. No C2+ hydrocarbons are detected. The catalyst con-
tacted with acetic acid (Fig. 10) leads to the formation of
methane and carbon dioxide, which appear simultaneously
at 550 K. This corresponds to the temperature of hydro-
gen consumption. A broad low-temperature peak of water
analogous to that observed with formic acid is also detected
along with small amounts of acetic acid. Neither ethane nor
propane molecules are detected. As can be seen, the TPH
spectra of formate and acetate surface species are very dif-
ferent from that of sample 4, indicating that the former
species are more likely not present in the reacting catalyst.
The methane TPH curve can be deconvoluted into four
Gaussian peaks a, b, c, and d at ca. 470, 580, 650, and 820 K,
respectively (Fig. 11). The corresponding species are de-
nominated , , , and . The ethane and propane profiles,
after deconvolution, lead to three peaks at ca. 470, 550, and
630 K; no C2+ are observed at 820 K. These observations
suggest that the species does not belong to the same fam-
FIG. 9. TPH profiles of sample 4 contacted with formic acid/He at
300 K: intensity in arbitrary units (a.u.) as a function of the TPH temper-
ature.
ily as the
, , and species; the latter, which leads to C2+
TPD experiments (Fig. 1) and which may be coincidental
with a methane evolution peak (644 K), is not well under-
stood.
hydrocarbons, are good candidates for being reaction inter-
mediates. The specificity of species has been confirmed by
the following analysis: The TPH experiment was stopped at
513, 606, 682, 873, and 973 K, and the saturation magnetiza-
tion of cobalt was subsequently measured after cooling the
system at 300 K under a flow of pure helium. An increase in
the magnetization was observed for the three first tempera-
ture intervals (0.7, 4.2, and 1.8% , respectively), whereas no
change was observed for the interval 682–973 K. This ob-
servation might be explained by assuming that the species
consists of graphitic carbon, which would not interact with
cobalt.
The possibility of formate and acetate formation in the
course of the reaction has been considered. After reduc-
tion, sample 4 was contacted at room temperature with a
mixture of helium saturated with formic or acetic acid, then
purged with pure helium and submitted to a TPH experi-
ment. With formic acid, the products detected were H2O,
CO2, CO, CH4, and H2 (Fig. 9). A CO2 peak is observed
at 485 K, a double peak of CO at 420 and 480 K, and a
methane peak at 515 K, related to a hydrogen consump-
tion peak at 515 K. A hydrogen formation peak at 480 K
might correspond to the CO2 and CO peaks at 485 and
FIG. 11. TPH profile of methane for sample 4 after 24 h of reaction
FIG. 10. TPH profiles of sample 4 contacted with acetic acid/He at
and decomposition of the methane TPH curve into four Gaussian peaks,
300 K: intensity in arbitrary units (a.u.) as a function of the TPH temper- a, b, c, and d, at ca. 470, 580, 650, and 820 K, respectively: intensity in
ature.
arbitrary units (a.u.) as a function of the TPH temperature.

Co/SiO2 CATALYSTS IN FISCHER–TROPSCH REACTION
113
Influence of Reaction Time
Figure 12 shows the variations in the normalized inten-
sity of each product with temperature for the three reaction
times (the intensity of the maximum peak is taken equal
to unity). As can be seen, the TPH curves obtained after
10 min of reaction are very different from those obtained
after 3.5 and 24 h; for example, a water production peak is
observed at high temperature after 10 min and is no longer
detected after 3.5 and 24 h. This peak does not coincide
with any carbon oxide production. Furthermore, it corre-
sponds to a small hydrogen consumption. This observation
might result from the reduction of a cobalt species not easily
reducible such as an anhydrous silicate or a phyllosilicate
produced by hydrothermal treatment of the catalyst result-
ing from the initial overheating in the presence of water
formed during the reaction. As reaction time increases, the
high-temperature reduction phenomenon disappears and
the reduction of the easily reducible species previously de-
scribed takes place. Variations in the carbon monoxide pro-
file also illustrate the change in catalyst characteristics oc-
curring in the course of the reaction.
FIG. 13. Variation of the peak intensity of methane (millimoles of
methane corresponding to the TPH peaks: a (+), b (᭞), c (2), and d (᭺))
with time.
the species) is different: the concentration of the species,
the most abundant species at short reaction times, first de-
creases strongly whereas those of the and species in-
crease. This strongly suggests that the species is the pre-
cursor of the other species. After a time in excess of 3.5 h,
the concentrations of , , and species stabilize.
Figure 13 shows the variations in the peak intensity of
methane with time. The species is not observed at short
reaction time. Its intensity increases with reaction time. This
observation is compatible with our hypothesis that the
species corresponds to the formation of graphitic deposits
accumulating in the catalyst. The evolution of the , , and
species (which lead to C2+ hydrocarbons in contrast with
Influence of Particle Size on TPH Data
Figure 14 illustrates the TPH profiles of samples 1–4 after
24 h of reaction. The water curve can be decomposed into
three Gaussian peaks at 440, 605, and 1120 K. The high-
temperature peak is not observed for sample 4, whereas
it is present for samples 1–3. This peak, which coincides
with hydrogen consumption, is observed for sample 4 after
10 min of reaction and has been assigned to the presence of
a cobalt compound not easily reducible such as a silicate or
a phyllosilicate. The CO profile goes through three maxima
at 480, 640, and 920 K and the CO2 curve yields four identi-
fiable peaks at 420, 515, 645, and 875 K. The methane curves
can be decomposed into four peaks regardless of the cobalt
dispersion at ca. 470, 580, 650, and 820 K. The first three
peaks are associated with ethane and propane formation,
in contrast with the 820 K peak. Figure 15 shows the vari-
ation in carbon coverage, Rx, associated with the different
carbon species present on the catalyst, with the diameter of
the cobalt particles. Rx is equal to the ratio of the number of
carbon atoms as calculated from TPH experiments to the
number of surface cobalt atoms. The subscript x denotes
the peak intensity (a, b, c, and d at 470, 580, 650, and 820 K,
respectively). As can be seen, an analogy between the vari-
ations in Rb and Rd is observed: the coverages of the and
species increase more or less continuously with D.
FIG. 12. TPH profiles of H2O, CO, CH4, CO2, C2H6, and H2 for
sample 4 after 10 min, 3.5 h, and 24 h of reaction: normalized intensity
as a function of the TPH temperature.

114
BARBIER ET AL.
DISCUSSION
EXAFS
EXAFSshowsthat Co in samples1, 4, and 5crystallizesin
the fcc crystal lattice as bulk Co. The average number of Co
neighbors in the consecutive coordination shells in the sam-
ples is lower than in the Co foil, indicating that clusters with
the fcc structure ofCo metal are formed. However, the aver-
age sizes of the clusters, deduced from the observed reduc-
tion in the average number of neighbors, are significantly
smaller than those obtained by magnetic measurements or
microscopy. Such observations on catalytic samples have
been previously reported by different authors (12, 18). An
explanation of the discrepancy, besides that cited earlier
(12) which employs specific models of disorder at the clus-
ter surface, can be given by an aggregate morphology of Co
particles.
It should be emphasized that the directly determined ex-
perimental parameter in EXAFS analysis is the average
number of neighbors. It is correlated linearly to the frac-
tion of surface atoms in a cluster, as can easily be shown for
large clusters: atoms in the core contribute with the full co-
ordination number of neighbors (12 in fcc lattice), whereas
those at the surface contribute with a deficient number (9
in fcc). Thus, the average number of (first) neighbors is a di-
rect measure of the specific surface of the dispersed metal,
and (should be) is directly related to the catalytic activity.
The conclusion is largely independent of the cluster shape
and size distribution. The value of the cluster diameter, on
the other side, depends critically on the assumptions of uni-
formity and globular shape. In the other two cases of simple
shapes, cylindrical rods and platelets, the extracted size pa-
rameter (diameter of rods and thickness of the platelets)
would differ for a small numerical factor (2/3 in fcc). How-
ever, it would invariably be the small shape parameter;
the average number of neighbors is largely insensitive to
the large size parameter, i.e., the length of the rods or the
width of the platelets. The conclusion can be extended to
aggregates of simple shapes (dendrites of globules, rods or
platelets) as long as the attachment area is small.
FIG. 14. TPH profiles of H2O, CO, CH4, CO2, C2H6, and H2 for sam-
ples 1–4 after 24 h of reaction: normalized intensity as a function of the
TPH temperature.
If globular clusters with perfect fcc structure are formed,
as assumed in the EXAFS model, then EXAFS is shown
to give reliable estimates of the average particle size (9,
10, 18). By contrast, larger aggregates composed of small
globular Co crystallites attached to each other by only a
small fraction of their surface give the large overall diam-
eters obtained by magnetic measurements or microscopy.
For example, samples 1 and 4 prepared in ammonia pos-
sess clusters with diameters of ca. 4.4 and 9.2 nm, respec-
tively while the corresponding globule sizes are 0.722 and
1.021 nm.
FIG. 15. Variations of the carbon coverage, Rx, associated with the
different carbon species present on the catalyst, with surface average di-
ameter (Ds). Rx is equal to the ratio of the number of carbon atoms as
calculated from TPH experiments to the number of surface cobalt atoms.
The subscript x denotes the peak identity (a (+), b (᭞), c (2), and d (᭹)
at 470, 580, 650, and 820 K, respectively).
Tight aggregates are also possible, whereby the small
globular clusters assume the role of domains in polycrys-
talline materials. The EXAFS estimate of the size refers
to the region of short-range order within a domain. A

Co/SiO2 CATALYSTS IN FISCHER–TROPSCH REACTION
115
similar model has been previously reported for Ni/SiO2 cat- demonstrate the same behavior (19): the selectivity toward
alysts(19) and MoS2 on oxidicor carbon carriers(12). There C2+ hydrocarbons increases with nickel particle size, then
is no direct relationship between the size of the domains stabilizes for diameters larger than ca. 6 nm.
and those of the large cobalt particles. It is also not clear,
however, what mechanism would produce that particular the parallel variations in intrinsic activity and chain growth
morphology. probability with cobalt particle size. More specifically, it has
Various hypotheses have been proposed to account for
Due to the ambiguities of EXAFS models we tried to been speculated that the CO coordination could be favored
relate the catalytic activity of the samples to particle di- on larger particles (2), thus increasing the CO dissociation
ameters and dispersion calculated from magnetic measure- rate and the intermediate concentration. Our TPH data
ments, which give results in good agreement with electron (Fig. 14) do not contradict this assumption: the position
microscopy.
of the high-temperature peak of CO evolution (probably
related to the highly coordinated species) shifts to lower
temperatures when cobalt particle size decreases from 9 to
4 nm. It cannot be overlooked, however, that in the case
of nickel catalysts that yield the same behavior as cobalt
catalysts, the ensemble model accounts for activity and se-
lectivity variations with particle size. These two hypotheses
are not mutually exclusive as the highly coordinated CO ad-
sorbed species probably requires an ensemble composed of
several adjacent metal atoms as an adsorption site.
Catalytic Activity
For practicable applications, catalysts yielding the high-
est conversion associated with the lowest weight of cobalt
are desirable. Our results (Fig. 5) show that, from this point
of view, the optimal diameter for cobalt particles is close to
6 nm. The corresponding rate of deactivation is, however,
large (Fig. 7). Finally, if one tries to reconcile the parame-
ters high specific activity, low rate of deactivation, and high
chain growth probability, our data show that Co/SiO2 cata-
lysts with D 9 nm are to be preferred in Fischer–Tropsch
synthesis.
Reacting Intermediate
Among the three carbon-containing species leading to
a mixture of methane and C2+ hydrocarbons detected by
TPH, the species seems to be related to the reacting in-
termediate; in fact the variations in its coverage with cobalt
particle size parallels that of the intrinsic activity. It is in-
ferred that the species may be identified with the reacting
intermediate of the reaction. The temperature of the peak
maximum for (550 K) does not coincide with that of the
oxygenates (H2O, CO, CO2), suggesting that this species
does not contain oxygen atoms ([CHx]n). Its hydrogenation
leads to the largest magnetic effect (increase in magneti-
zation by 4.2% ), confirming that this intermediate is inter-
acting with the metal cobalt phase. Its coverage order of
magnitude (0.2–0.3) is comparable to that observed over a
similar nickel catalyst by transient kinetics (0.1–0.3) (21). It
could be argued that the species, which leads mostly to
methane in TPH experiments, cannot be the reacting inter-
mediate in the Fischer–Tropsch reaction, on the basis that
the latter gives heavy hydrocarbons. However, this argu-
ment can be rebutted considering that the Fischer–Tropsch
reaction and TPH occur under very different experimental
conditions. TPH, which is performed at increasing temper-
atures in the absence of CO, might favor the hydrogenolysis
of Cn species into methane, thus explaining the difference
in the nature of the products with the Fischer–Tropsch re-
action.
Size Sensitivity
The trend toward size insensitivity we observe for the
intrinsic activity for diameters larger than 6–9 nm (Fig. 5)
agrees well with the data of Ho et al. (1) at atmospheric
pressure, Iglesia et al. (3), and Azib (5) at P = 2 MPa, who
reported no important variations for diameters in excess of
5, 10, and 9 nm, respectively. Over the low-diameter range,
our results are also in qualitative agreement with those of
Reuel et al. (2), who have reported an increase in activity
for cobalt particle diameters larger than 1 nm (dispersion
86% ). The use of chemisorption techniques in the latter
case casts some doubt on the quantitative aspects of the
size determination, however. Furthermore, these authors
have used partially reduced catalysts, so their conclusions
are not completely unambiguous. In our case, the catalysts
are initially fully reduced so that the hypothesis of a possi-
ble extent of reduction effect can be discarded to account
for the size sensitivity observed. Finally, it can be under-
lined that our results are very similar to those obtained by
Dalmon and Martin (19) on nickel catalysts, where a steep
decrease in the intrinsic activity was reported for diameters
lower than 6 nm.
The variations in chain growth probability observed in
this work (Fig. 6) are also in agreement with literature data:
generally, decreases with dispersion (4–6, 20), whereas a
size insensitivity is observed for large particles in the disper-
sion range 0.45–9.5% (3). Our study provides a quantita-
Deactivation
The same type of analogy might be found between the
tive determination of the critical diameter separating these deactivation rate and the c coverage related to the
two zones, i.e., 6 nm. The size sensitivity of Co/SiO2 cata- species (Figs. 7 and 15): both parameters first increase, go
lysts is similar to that observed for Ni/SiO2 catalysts which through a maximum, and then decrease as cobalt particle

116
BARBIER ET AL.
size increases. The parallel, however, is much less quantita-
In the course of the reaction, neither formates nor ac-
tive than in the case of the intrinsic activity and the coverage etates are detected in the catalyst. Four carbon-containing
of the species; in fact, the difference between the extreme species are observed. The most reactive toward hydrogen,
values of the c coverage does not exceed 50% , whereas the
rate of deactivation varies by a factor of 7 when D increases decreases with time, whereas the second, , has been tenta-
from 5.5 to 9 nm. tively assigned to the active phase of the reaction on the
, is abundant at low reaction time and its concentration
A better clue to identify the main cause of deactivation basis of its parallel variation with intrinsic activity. This
is provided by careful examination of the TPH curves of intermediate, which is strongly interacting with the cobalt
water evolving at high temperature, associated with a high- phase, does not contain oxygen atoms. The species might
temperature hydrogen consumption peak, revealing the be associated with graphitic residue formation, whereas no
presence of an oxidized cobalt phase not easily reducible specific role has been attributed to the species.
(Fig. 14). It is striking to observe that after 24 h of reac-
tion, the most stable catalyst (sample 4, Ds = 9.2 nm) does
ACKNOWLEDGMENTS
not contain this phase, in contrast with samples 1 to 3, the
The EXAFS study is supported by Internationales Buero des BMBF,
Germany, and Ministry of Sciences and Technology, Slovenia. Expert ad-
vice on the EXAFS 2 beamline operation of M. Tischer from HASYLAB
is gratefully acknowledged.
least stable catalysts. It can be inferred that deactivation is
probably due to a loss of catalytically active cobalt, which
is oxidized to irreducible cobalt silicate or phyllosilicate in
the course of the reaction (22, 24).
REFERENCES
CONCLUSION
1. Ho, S. W., Houalla, M., and Hercules, D. M., J. Phys. Chem. 94, 6396
(1990).
The formation of Co clusters with the fcc crystal struc-
ture in Co/SiO2 catalysts is definitely established by EX-
AFS. Estimated average sizes of the clusters deduced from
EXAFS data are significantly smaller than those obtained
by magnetic measurements or TEM. The observed dis-
crepancy suggests a complex morphology for Co particles.
A dendritic or polycrystalline structure of large Co parti-
cles, composed of several small individual Co crystallites,
is proposed, a situation already met in the case of some
Ni/SiO2 catalysts. Hydrogen evolution from TPH experi-
ments appears to be so complex that it casts some doubt on
the reliability of chemisorption techniques for the estima-
tion of cobalt dispersion. The ambiguities of EXAFS and
chemisorption methods have led us to use diameters and
dispersions calculated from magnetic measurements, which
give results in good agreement with electron microscopy, in
the presentation of our catalytic results.
The size sensitivity of the Fischer–Tropsch reaction has
been confirmed: intrinsic activity (per unit area) and chain
growth probability first increase, then stabilize at increasing
particle size. Our data provide a quantitative determination
of the critical diameter separating these two zones, i.e., ca.
6 nm. A topological hypothesis (high index planes could fa-
vor highly coordinated chemisorbed CO) has already been
put forward to account for this observation: this possible
explanation is not exclusive of that based on the ensem-
ble model proposed for the case of Ni/SiO2 catalysts which
yield the same behavior. The rate of deactivation is also
sensitive to cobalt particle size. TPH experiments indicate
that deactivation may be due to a loss of catalytically active
cobalt, which is transformed into irreducible cobalt com-
pounds under our reacting conditions.
2. Reuel, R. C., and Bartholomew, C. H., J. Catal. 85, 78 (1984).
3. Iglesia, E., Soled, S. L., and Fiato, R. A., J. Catal. 137, 212 (1992).
4. Bartholomew, C. H., Catal. Lett. 7, 27 (1990).
5. Azib, H., Thesis, University P. and M. Curie, Paris VI, 1996.
6. Reuel, R. C., and Bartholomew, C. H., J. Catal. 85, 63 (1984).
7. Barbier, A., Hanif, A., Dalmon, J. A., and Martin, G. A., Appl. Catal.
A 168, 333 (1998); Barbier, A., Brun-Pereira, E., and Martin, G. A.,
Catal. Lett. 45, 221 (1999).
8. Rehr, J. J., and Albers, R. C., Rev. Mod. Phys. 72, 621 (2000).
9. Frenkel, A. I., J. Synchrotron Radiat. 6, 293 (1999).
10. Borowski, M., J. Phys. IV. 7(C2), 259 (1997).
11. kakar, S., Bjorneholm, O., Weigelt, J., De Castro, A. R. B., Troeger,
L., Frahm, R., Moeller, T., Knop, A., and Ruehl, E., Phys. Rev. Lett.
78, 1675 (1997).
12. Shido, T., and Prins, R., J. Phys. Chem. B 102, 8426 (1998).
13. Bensaddik, A., Ernst, B., Garin, F., Maire, G., and Kiennemann, A.,
J. Phys. IV 6(C4), 545 (1996).
14. Jorda, E., Tuel, A., Teissier, R., and Kervennal, J., J. Chem. Soc. Chem.
Commun., 1775 (1995).
15. Rher, J. J., Albers, R. C., and Zabinsky, S. I., Phys. Rev. Lett. 69, 3397
(1992).
16. Stern, E. A., Newville, M., Ravel, B., Yacoby, Y., and Haskel, D.,
Physica B, 117 (1995).
17. D’Acapito, F., Cattaruzza, E., Gonella, F., Mobilio, S., Regnard, J. R.,
and Spizzo, F. J., J. Phys. IV 7(C2), 1107 (1997).
18. Montano, P. A., Shenoy, G. K., Alp, E. E., Shulze, W., and Urban,
J., Phys. Rev. Lett. 56, 2076 (1986).
19. Dalmon, J. A., and Martin, G. A., J. Catal. 84, 45 (1983).
20. Van de Loosdrecht, J., Van der Haar, M., Van der Kraan, A. M., Van
Dillen, A. J., and Geus, J. W., Appl. Catal. A 150, 365 (1997).
21. Agnelli, M., Swaan, H. M., Marquez Alvarez, C., Martin, G. A., and
Mirodatos, C., J. Catal. 175, 117 (1998).
22. Coulter, K. E., and Sault, A. G., J. Catal. 154, 56 (1995).
23. Kogelbauer, A., Weber, J. C., and Goodwin J. G., Jr., Catal. Lett. 34,
259 (1995).
24. Van Berge, P. J., Van de Loosdrecht, J., Barradasa, S., and Van der
Kraam, A. M., Catal. Today 58, 321 (2000).