On the nature of different iron sites and their catalytic role in Fe-ZSM-5 DeNOx catalysts: New insights by a combined EPR and UV/VIS spectroscopic approach

Article abstract of DOI:10.1016/j.jcat.2004.08.003

Fe-ZSM-5 DeNO_x catalysts prepared from H-ZSM-5 by different ion-exchange procedures were analyzed ex situ after synthesis, calcination, and use in catalysis as in situ during calcination. The results were correlated with the catalytic behavior of these materials in SCR of NO by isobutane or ammonia. The SCR by ammonia was catalyzed by different entities (mononuclear Fe sites, Fe_xO_y oligomers, surface of iron oxide particles). Mononuclear Fe sites were involved in the SCR with isobutane. Fe_xO_y cluster formation was favored by moisture remaining in the zeolites after washing. A participation of isolated sites was inferred from the observation that during the use of the catalysts in the SCR reaction on a timescale where no deactivation was noted, oligomeric clusters aggregated and achieved a more perfect order while the amount of isolated sites remained constant.

Full text of DOI:10.1016/j.jcat.2004.08.003

Journal of Catalysis 227 (2004) 384–397
www.elsevier.com/locate/jcat
On the nature of different iron sites and their catalytic role in Fe-ZSM-5
DeNOx catalysts: new insights by a combined EPR and UV/VIS
spectroscopic approach
M. Santhosh Kumar a, M. Schwidder b, W. Grünert b, A. Brückner a,∗
a
Institut für Angewandte Chemie Berlin-Adlershof e. V., Richard-Willstätter-Str. 12, D-12489 Berlin, Germany
b
Lehrstuhl für Technische Chemie, Ruhr-Universität Bochum, D-44780 Bochum, Germany
Received 28 April 2004; revised 6 July 2004; accepted 4 August 2004
Available online 3 September 2004
Abstract
Fe-ZSM-5 DeNO catalysts prepared from H-ZSM-5 by different ion-exchange procedures have been analyzed by EPR and UV/VIS
x
diffuse reflectance spectroscopy (DRS) ex situ after synthesis, calcination, and use in catalysis as well as in situ during calcination. The
results have been correlated with the catalytic behavior of these materials in the selective catalytic reduction (SCR) of NO by isobutane or
ammonia. In comparison to previous studies of the same samples by XAFS, XRD, XPS, TPR, and Mössbauer spectroscopy, the combination
of EPR and UV/VIS-DRS was more sensitive for distinguishing between different types of isolated Fe species as well as Fe O aggregates
x
y
of different size (oligonuclear clusters or large particles). It was found that aggregated species are formed at the expense of mononuclear Fe
sites upon calcination at 873 K, and that aggregate formation is slightly favored by calcination with higher heating rates as well as by high
Si/Al ratios of the parent H-ZSM-5. Use in SCR of NO leads to further growth and restructuring of Fe O clusters. From the comparison of
x
y
structural and catalytic properties of different Fe-ZSM-5 catalysts it can be concluded that the SCR of NO by NH is catalyzed by different
3
entities (mononuclear Fe sites, Fe O oligomers, surface of iron oxide particles). The results suggest that mononuclear Fe sites are also
x
y
involved in the SCR with isobutane. Clustered sites, which may contribute to SCR with isobutane as well, appear to cause nonselective
oxidation of the reductant (isobutane or ammonia) at higher temperatures.
2004 Elsevier Inc. All rights reserved.
3+
Keywords: EPR; UV/VIS-DRS; Fe-ZSM-5 DeNO catalysts; Fe single sites; Iron oxide clusters
x
1. Introduction
ZSM-5 matrix by Fe2+ or Fe3+ ions are nowadays largely
disregarded despite reports of considerable achievements
with special techniques [5,7,8], because they most likely
involve iron oxo- or hydroxo cations that can undergo com-
plex chemical transformation during subsequent calcination
and are believed to lead to highly heterogeneous materials
with poor reproducibility. Instead, attention was focused on
overexchangedFe-ZSM-5 obtained by chemical vapor depo-
sition (CVD) of FeCl3 into the pores of H-ZSM-5 followed
by subsequent washing and calcination steps [5]. The ma-
terial was prepared by many groups, which reported high
and reproducible NO reduction activities with the reductant
isobutane [5,9–11], but also outstanding activities with the
reductant ammonia [4,12,13].
In recent years, MFI-type zeolites containing nonframe-
work iron species, such as Fe-ZSM-5, have been recognized
as highly efficient catalysts for a number of reactions, e.g.,
selective oxidation of benzene to phenol [1,2], N2O decom-
position [3], and selective catalytic reduction (SCR) of NOx
by NH3 [4] and hydrocarbons [5,6]. In the preparation of Fe-
ZSM-5 catalysts, traditional aqueous exchange techniques
aiming at the replacement of H+ or Na+ cations of the
*
Corresponding author. Institut für Angewandte Chemie Berlin-Adlers-
hof e. V., PO Box 96 11 56, D-12474 Berlin. Fax: +49 30 6392 4454.
E-mail address: brueckner@aca-berlin.de (A. Brückner).
0021-9517/$ – see front matter
doi:10.1016/j.jcat.2004.08.003
2004 Elsevier Inc. All rights reserved.

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
385
In this preparation, zeolite Brønsted sites are completely
consumed in the CVD step but reappear partly after calci-
nation. Together with a recourse on enzymatic models, this
has inspired the view that the active sites responsible for the
high SCR activities are binuclear intrazeolite Fe oxo com-
plexes, which has been supported by several EXAFS inves-
tigations [14–17]. Other groups concluded from their char-
acterization studies that their Fe-ZSM-5 samples, which ex-
hibited comparable SCR activities, contain a multitude of Fe
species, among them isolated sites, clusters of a wide range
of nuclearities and large oxide aggregates [10,11]. There-
fore, alternative suggestions about the active site structure of
Fe-ZSM-5 SCR catalysts have been also made. They include
Fe4O4 clusters [18], isolated mononuclear sites [19], or all
oligomeric clusters with low nuclearities including mononu-
clear sites [10,20]. Large oxide particles are believed to be
rather inactive [10,21].
The heterogeneity of the Fe site structure in Fe-ZSM-5
prepared via the CVD method has been suggested on the
basis of a multitechnique study comprising XRD, EXAFS,
Mössbauer spectroscopy, TPR, FTIR, and XPS [10]. In this
study, the preparation was applied to parent zeolites of dif-
ferent Si/Al ratio, and different washing and calcination pro-
tocols were used to create catalysts with a different degree
of site isolation. A sample made by solid-state ion exchange
(SSIE), which contained large Fe2O3 aggregates outside the
pores along with other Fe species, was also included as well
as a material prepared by a novel mechanochemical route,
which, based on EXAFS results, was believed to contain ex-
clusively Fe3+ single sites [20]. It turned out, however, that
a reliable identification of the different Fe species coexisting
in these materials and, in particular, their relative quantities,
is not straightforward because the sensitivity of the tech-
niques for the various types of Fe species differs. Thus, it
was found that EXAFS is most sensitive for highly disperse
Fe sites while the scattering contribution from higher shells,
which is expected to indicate the presence of FexOy clus-
ters, was near the detection limit or hardly significant for all
samples except the one prepared by SSIE, in which crys-
talline, probably extrazeolite α-Fe2O3 was also detected by
XRD. In contrast, Mössbauer and TPR measurements in-
dicated a much larger degree of clustering than was found
by EXAFS. A major goal of the present study is to clarify
these contradictions and to give more insight into the struc-
tures of Fe oxo sites formed by the preparations typically
employed in the literature, in particular the CVD technique.
This was achieved by the use of additional techniques which
are able to distinguish between isolated Fe species of differ-
ent structures on the one hand and between FexOy clusters
of different nuclearity on the other hand. EPR first of all and
also UV/VIS spectroscopy are suitable techniques for this
purpose.
sis of the mutual magnetic interactions of the Fe sites [26].
UV/VIS spectroscopy, on the other hand, is especially sensi-
tive to charge-transfer(CT) bands of Fe3+, the wavelength of
which depends on the coordination number and the degree of
aggregation [27]. We have therefore reexamined the samples
studied in [10] and [20] by these methods with the aim to
better differentiate the Fe sites present and to detect changes
in their structure caused by different preparation conditions
(mode of Fe introduction, Si/Al ratio of the parent H-ZSM-
5, calcination, and washing protocols) and by use in the SCR
of NOx than it was possible with the methods used in the
previous study. It will be confirmed that Fe-ZSM-5 cata-
lysts with high Fe content—even if prepared by the CVD
technique—contain iron in a wide variety of species. While
this is an unfavorable situation for identifying active sites
for any reaction, it will be shown that the qualitative trends
in the structural and catalytic data obtained support the view
that mononuclear Fe ions are involved in the SCR of NO by
isobutane and by ammonia. A study with catalysts prepared
by a technique that allows highly disperse Fe sites with a
better control of site distribution to be produced will be pre-
sented in the near future.
2. Experimental
The preparation of the samples has been described in
detail in [10,20]. Therefore, only a brief summary will be
given here, and the sample code used to label the samples
with reference to the preparation details will be explained.
Two ZSM-5 materials with different Si/Al ratios (≈ 14 and
≈ 40, the latter with high density of structure defects, e.g.,
silanol nests, labeled A and B, respectively) have been em-
ployed. The first one (A) was used with all preparations,
the other one only for comparative purposes. Fe ions have
been introduced into the parent H-ZSM-5 matrices by three
different methods: (i) Chemical vapor deposition using an-
hydrous FeCl3 followed by washing with 1 or 10 L water
per 5 g catalyst (W1 and W10, respectively) and calcina-
tion in air at 873 K with heating rates of 0.5 or 5 K/min
(C0.5 and C5, respectively), (ii) solid-state ion exchange
with FeCl3 × 6H2O—different from the procedure outlined
in [10] with subsequent washing, which leads to a significant
decrease of the iron content in the case of the SSIE prepara-
tion, and (iii) treatment by a mechanochemical route (MR)
that implies intense grinding of the parent H-ZSM-5 with
FeCl3 × 6H2O, followed by 2–3 short-time washing steps
(0.5 L water per 2 g catalyst). This material was studied in
the present work starting from the uncalcined state, while the
usual calcination conditions were 873 K, air, 1 h as with the
other materials [20].
The label is composed of the symbols for the ZSM-5
type, Fe introduction route, washing intensity, and heating
rate during calcination. Thus, A(CVD, W1, C0.5) means a
material in which Fe was introduced via CVD into ZSM-
5 of type A, washed with 1 L water per 5 g catalyst, and
EPR spectroscopy is a powerful tool for identifying iso-
lated Fe3+ species of different coordination geometries by
the position of their signals [2,8,21–25] as well as FexOy
cluster species of different degrees of aggregation by analy-

386
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
calcined with a heating rate of 0.5 K/min. Missing sym-
bols mean that the corresponding step has been omitted (i.e.,
A(CVD, W1) is A(CVD, W1, Cx) before calcination with
a ramp of x K/min). The Fe content of the samples deter-
mined from the X-ray absorption step height in the XAFS
spectra ([10,20], validated by ICP analyses of selected sam-
ples), amounted to 0.5 wt% for A(MR), 5.4 wt% for A(CVD,
W1), 5.0 wt% for A(CVD, W10), 5.2 wt% for A(SSIE,
W1), and 2.6 wt% for B(CVD, W1). It has been recently
reported that the introduction of iron via FeCl3 may plug
pores in the parent ZSM-5 [28]. Indeed, in our samples the
pore volume of the parent H-ZSM-5 zeolite decreased from
0.15 to 0.105 cm3/g for A(CVD, W1, C5), A(CVD, W10,
C5), and, surprisingly, A(MR, C5), while the pore volume
of B(CVD,W1, C5) was 0.084 cm3/g, and that of A(SSIE,
W1, C5) 0.064 cm3/g.
UV/VIS-DRS measurements were performed with a
Cary 400 spectrometer (Varian) equipped with a diffuse
reflectance accessory (praying mantis, Harrick). To re-
duce light absorption, samples were diluted by α-Al2O3
(calcined for 4 h at 1473 K) in a ratio of 1:10. Spec-
tra were measured in reflectance mode and converted into
the Kubelka–Munk function F(R) which is proportional
to the absorption coefficient for sufficiently low F(R) val-
ues. For a proper comparison of small variations of the
shape of different experimental spectra, the F(R) values
of the latter were normalized, i.e., multiplied by a certain
constant factor so as to obtain F(R) = 1 for the max-
imum of the spectrum. For in situ experiments, a heat-
able reaction chamber (Harrick) equipped with a temper-
ature programmer (Eurotherm) and a gas-dosing system
containing mass-flow controllers (Bronkhorst) was used.
Deconvolution of the UV/VIS spectra into subbands was
performed by the computer program GRAMS/32 (Galac-
tic).
EPR spectra in the X-band (ν ≈ 9.5 GHz) were recorded
with the cw spectrometer ELEXSYS 500-10/12 (Bruker)
using a microwave power of 6.3 mW, a modulation fre-
quency of 100 kHz, and a modulation amplitude of 0.5 mT.
The magnetic field was measured with respect to the stan-
dard 2,2-diphenyl-1-picrylhydrazyl hydrate (DPPH). For
temperature-dependent measurements in the range from 90
to 293 K, a commercial variable temperature control unit
(Bruker) and a conventional EPR sample tube was used,
while in situ EPR measurements during calcination in the
range 293 < T /K < 773 were performed in a homemade
EPR flow reactor [29].
The selective reduction of NO with isobutane or am-
monia was studied in a catalytic microflow reactor with
a product analysis scheme that combined calibrated mass
spectrometry, gas chromatography, and nondispersive IR
photometry (NH3). Feed gases containing 1000 ppm NO,
1000 ppm reductant (isobutane or ammonia), 2% O2 in He
were charged onto the catalyst at 30,000 h−1 for isobutane-
SCR and 750,000 h−1 for NH3-SCR. Generally, the catalytic
runs were started with a thermal activation and stabilization
treatment of the catalysts in flowing He at 550 ◦C (isobutane)
or at 600 ◦C (NH3). The activities were measured from the
higher to the lower reaction temperatures. Under our experi-
mental conditions, the only reaction product of NO observed
in the limits of experimental accuracy was nitrogen; i.e., the
NO conversions given are equal to N2 yields.
3. Results and discussion
3.1. Catalytic activity
In Fig. 1, NO conversions achieved with the various Fe-
ZSM-5 preparations and of the parent H-ZSM-5(A) are com-
Fig. 1. NO conversions in the selective catalytic reduction of NO with isobutane (a) or NH (b) over different Fe-ZSM-5 catalysts. Feed composition: 1000 ppm
3
−1
−1
NO, 1000 ppm isobutane (or NH ), 2% O in He, at 30,000 h (a) or 750,000 h (b). ", A(SSIE, W1, C5); E, A(CVD, W10, C0.5); , A(CVD, W1,
3
2
C5); ꢀ A(MR, C5); , B(CVD, W1, C5); +×, H-ZSM-5(A).

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
387
pared. Some of the data from isobutane-SCR have been al-
Table 1
Normalized reaction rates, r /Fe, for the SCR of NO with isobutane or
NH over Fe-ZSM-5 of different preparations
3
NO
ready shown in [10]. A(MR, C5) is a new preparation the
activity of which even exceeds that of the material described
in [20]. As known from the literature, the activity of the cat-
alysts with the ammonia reductant was found to be much
larger than with isobutane [4,12,13].
a
a
Preparation
Isobutane-SCR
NH -SCR
3
4
−1
4
−1
10 r /Fe, s at 598 K 10 r /Fe, s at 573 K
NO
NO
A(CVD, W1, C5)
A(CVD, W10, C0.5) 4.5
A(SSIE, W1, C5)
B(CVD, W1, C5)
A(MR, C5)
4.1
42.2
41.4
42.6
36.7
79.1
Three different types of behavior can be identified for
each reductant, but the details differ considerably. Cata-
lysts prepared via gas-phase transport of FeCl3 into H-ZSM-
5(A) (A(SSIE, W1, C5), A(CVD, W1, C5), and A(CVD,
W10, C0.5)) exhibit rather similar catalytic properties. In
isobutane-SCR (Fig. 1a), the members of this reference
group are hard to distinguish while in NH3-SCR (Fig. 1b),
the highest activity is obtained with A(SSIE, W1, C5). It
should be noted that pore plugging, which is obvious with
all catalysts from the pore volume data (see Experimental) is
most severe with A(SSIE, W1, C5); hence, significant parts
of the iron present may be included and not able to partic-
ipate in catalysis. At high temperatures, the catalysts of the
reference group exhibit remarkable differences in NH3-SCR.
The remaining catalysts show a contrary behavior relative
to this group. The material obtained by CVD of FeCl3 into
the defective ZSM-5 matrix B has lower activities with both
reductants, but while its activity is very poor with isobutane
it is still comparable with the other catalysts with ammonia
where it achieves approximately half the conversions with
half the Fe content. On the other hand, very high NO con-
versions are achieved with A(MR, C5), although at higher
temperatures—at low temperatures, this catalyst falls short
of the reference group as well. With ammonia, A(MR, C5)
never achieves higher NO conversions than the remaining
catalysts in accordance with the rule that a lower Fe content
causes lower conversions. It should be noted, however, that
the Fe content of A(MR, C5) is only 20% of that in B(CVD,
3.7
2.2
b
25.6
a
For conditions see legend to Fig. 1.
From interpolated NO conversion.
b
W1, C5) and 10% of that in the reference-group catalysts.
Therefore, a comparison on the basis of normalized reaction
rates rNO/Fe (reaction rate related to the total Fe content)
is interesting. Table 1 reports such data for temperatures
around 600 K (i.e., the temperature of peak NO conversion
over the reference-group catalysts in isobutane-SCR). Al-
though these data must be treated with some caution because
they have been derived from measurements under integral
conditions, they show that the intrinsic activity of the Fe sites
is the highest also under these conditions. This is, in par-
ticular, true for isobutane-SCR while the advantage is less
pronounced for NH3-SCR.
3.2. Structural characterization
3.2.1. Assignment of EPR signals
EPR spectra of representative samples (A(SSIE, W1),
A(CVD, W1, C0.5), and A(MR)), which were recorded at
different temperatures, are displayed in Fig. 2. The spectra
of all samples show three signals at effective g values of
gꢀ ≈ 6, gꢀ ≈ 4.3 and gꢀ ≈ 2. Such signals were frequently
detected in Fe-ZSM-5 [22,23,30,31] but also with Fe3+ ions
Fig. 2. EPR spectra of hydrated samples (a) A(SSIE, W1), (b) A(CVD, W1, C0.5), (c) A(MR)—uncalcined and (d) parent H-ZSM-5(A) during heating in air
flow. Spectra of A(MR) and parent H-ZSM-5(A) are multiplied by a factor of 2.

388
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
in other oxide matrices. However, their assignment is by no
means straightforward. The number and position of EPR
transitions for Fe3+ ions observable in a powder spectrum
depends sensitively on the local crystal field symmetry of
these sites (reflected by the magnitude of the zero field split-
ting parameters D and E) and possible mꢀagnetic interactions
between them. Signals at gꢀ ≈ 4.3 and g ꢁ 6 arise from the
|−1/2 >↔> 1/2| transition of isolated Fe3+ sites in strong
rhombic (D ꢁ hν, E/D = 1/3, gꢀ ≈ 4.3) or axial distor-
tion (D ꢁ hν, E = 0, gꢀ ≈ 6) when the zero field splitting is
large in comparison to the microwave energy hν [23,24,32].
Tꢀhis implies that an Fe3+ species giving rise to a line at
g ≈ 4.3 is more strongly distorted than an Fe3+ site rep-
resented by a signal at gꢀ ≈ 6 due to the difference in the
magnitude of E.
of small oligonuclear FexOy clusters. It is rather unlikely,
that a dimer Fe–O–Fe species can give rise to a gꢀ ≈ 2 signal
with the observed temperature dependence. The existence
of those moieties is discussed by several authors [14–17]
in comparison to a similar species in the enzyme methane
monooxygenase (MMO). However, it has to be noted, that in
oxidized MMO those Fe3+–O–Fe3+ dimers are EPR silent
due to spin pairing (S = 0). In the partially reduced MMO
a signal with gꢀ values of 1.94, 1.86, and 1.75 was observed
at 8 K for an oxo-bridged binuclear iron cluster and in the
completely reduced MMO a single line at gꢀ ≈ 16 has been
recorded at 8 K which was assigned to a dimer composed of
two ferromagnetically coupled high-spin Fe2+ centers [33].
None of those signals is observed in the Fe-ZSM-5 sam-
ples studied in this work. Therefore, we think that the pres-
ence of an oxo-bridged Fe dimer similar to that in MMO is
highly questionable in our Fe-ZSM-5, although it cannot be
rigorously excluded. Theoreticalꢀly, a Fe3+–O–Fe3+ dimer
species could contribute to the g ≈ 2 signal when its Neel
temperature is exceeded.
In zeolites, the line at gꢀ ≈ 4.3 is frequently assigned to
Fe3+ sites incorporated in tetrahedral framework positions
while a line at gꢀ ≈ 6 is assigned to isolated Fe3+ species in
higher coordination numbers [23,30,31]. However, it must
be stressed, that just from the signal position alone it is not
possible to conclude whether the respective Fe ions are oc-
tathedrally or tetrahedrally coordinated since the signal po-
sition is governed by the magnitude of D and E, i.e., the
extent of distortion of the Fe coordination. This distortion
can arise from both tetrahedral and octahedral coordination.
To draw cꢀonclusions on the number of ligands associated
with the g ≈ 4.3 and 6 signals, additional aspects must be
considered which are discussed in the following sections.
Given the preparation techniques, Fe-ZSM-5 samples
studied in this work are not expected to contain Fe3+ in
framework positions. However, XAFS measurements per-
formed with hydrated Fe-ZSM-5 A(CVD, W1, C0.5) and
A(MR, C5) samples in ambient atmosphere suggested that
the latter do contain a certain amount of isolated, tetrahe-
drally coordinated Fe3+ ions [10,2ꢀ0]. Moreover, as will be
shown below, an increase of the g ≈ 4.3 signal at the ex-
pense of gꢀ ≈ 6 and gꢀ ≈ 2 lines can be induced by dehydra-
tion, implying that the loss of water ligands leads to a lower
coordination number. Therefore, we assign the EPR signal
at gꢀ ≈ 4.3 to tetrahedrally coordinated Fe3+ species and the
line at gꢀ ≈ 6 to isolated (presumably higher coordinated)
Fe3+ species in less distorted extraframework positions.
Signals at gꢀ ≈ 2 can arise either from isolated Fe3+ in
high symmetry (D, E ≈ 0) or from FexOy clusters in which
magnetic interactions between the Fe3+ ions average out
the zero field splitting. For isolated, highly symmetric Fe3+
species, the signal intensity should follow the Curie–Weiss
law, i.e., I ∼ 1/T . It appears that both types of signal can be
observed in the spectra: A narrow signal found at gꢀ ≈ 2 in
A(MR) below 250 K (Fig. 2c) disappears upon heating while
the broad superimposed signal, which is also present in the
spectra of the other samples, narrows and increases with
rising temperature. This points to antiferromagnetic interac-
tions that collapse upon heating and suggests that the broad
line at gꢀ ≈ 2 arises rather from neighboring than from iso-
lated Fe3+ sites. Most probably such species are constituents
3.2.2. Assignment of UV/VIS signals
The UV/VIS spectra of Fe-ZSM-5 zeolites presented in
this work have been deconvoluted into subbands to facilitate
assignment to different Fe species. In principle, two ligand-
to-metal charge-transfer (CT) transitions, t1 → t2 and t1 →
e, are to be expected for a Fe3+ ion [34]. For isolated Fe3+
sites they give rise to bands below 300 nm, whereby their
particular position depends on the number of ligands. Thus,
CT bands of isomorphously incorporated tetrahedrally coor-
dinated Fe3+ ions have been observed at 215 and 241 nm in
Fe–silicalite [27] while a band at 278 nm is detected for iso-
lated octahedral Fe3+ sites in Al2O3 [35]. Although a clear
discrimination of CT bands of isolated Fe3+ sites in tetra-
hedral and higher coordination is not straightforward due
to their similar wavelength range, these values suggest that
CT bands of Fe3+ are red-shifted with increasing number of
coordinating oxygen ligands. The same trend has been ob-
served accordingly also for V5+ species [36].
In the UV/VIS spectrum of sample A(MR) two intense
bands appear at 228 and 290 nm (Fig. 3b). Previous EXAFS
measurements revealed that this sample is dominated by iso-
lated Fe3+ sites with a mean coordination number between
4 and 6, suggesting that these Fe species are in both tetrahe-
dral and higher coordination [20]. In agreement with these
results we assign the two strong bands in Fig. 3b to isolated
Fe3+ sites in tetrahedral (228 nm) and higher coordination
(290 nm). For spectra deconvolution, only one CT band has
been used for each of the two isolated Fe3+ species, since
separate bands for the two possible CT transitions of each
type of Fe3+ species (t1 → t2 and t1 → e) are not resolved
in the experimental spectrum.
According to the literature, CT bands between 300 and
400 nm are assigned to octahedral Fe3+ in small oligomeric
FexOy clusters [27] while bands above 450 nm arise from
larger Fe2O3 particles as can be seen, too, from the spec-

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
389
Fig. 3. UV/VIS diffuse reflectance spectra of hydrated as-prepared samples, compared with α-Fe O ; experimental spectra, thick solid lines; deconvoluted
2
3
3+
subbands, thin lines; assignments: - - - isolated Fe , — small oligomeric Fe O moieties, ··· extended Fe O -like clusters, (a) A(CVD, W1), (b) A(MR),
x
y
2
3
(c) A(SSIE, W1), (d) α-Fe O .
2
3
trum of the reference sample α-Fe2O3 (Fig. 3d). In general,
a t1 → t2 and a t1 → e transition should be expected, too, for
Fe3+ in cluster-like FexOy species. However, light absorp-
tion in the spectra of Figs. 3a and c above 300 nm occurs
in a very broad range, suggesting the superposition of those
bands for a variety of slightly different small oligonuclear
FexOy clusters and larger Fe2O3 species. Thus, for spectra
deconvolution the lowest possible number of subbands in the
range above 300 nm has been used that was needed to obtain
a satisfactory fit of the experimental spectrum. This proce-
dure is regarded to be acceptable, although a deconvolution
of a broad and poorly structured experimental spectrum into
subbands by mathematical means is always arbitrary to a cer-
tain extent. The subbands above 300 nm must be understood
in terms of reflecting a certain distribution of slightly dif-
ferent cluster geometries rather than representing a certain
number of different individual cluster species. The percent-
age of the subbands with respect to the total area of the
experimental spectrum has been multiplied with the overall
Fe content (accurately determined by both EXAFS and ICP-
OES) to obtain a rough estimate of the percentage of the
different Fe species in the samples (Table 2). Being aware
that the obtained values in Table 2 are inaccurate to a certain
degree due to the deconvolution procedure, they are never-
theless regarded to be helpful for a comparison of different
samples.
Table 2
Percentage of the area of the subbands (I at λ < 300 nm, I at 300 < λ <
1
2
400 nm, and I at λ > 400 nm) derived by deconvolution of the UV/VIS-
3
DRS spectra (see Figs. 3, 4, 6, and 7) and corresponding Fe percentage
derived from total Fe content determined by XAFS [10,20]
a
b
c
I
Catalyst
I
Fe
1
I
Fe
2
Fe
3
1
2
3
(%) (wt%) (%) (wt%) (%) (wt%)
A(MR)
A(MR, C5)
A(CVD,W1)
A(CVD,W1,C0.5)
A(CVD,W1,C0.5) used
A(CVD,W1,C5)
91
83
47
27
25
27
22
46
0.45
0.41
2.53
1.46
1.35
1.46
1.19
2.30
1.35
1.66
1.56
0.73
0.68
9
17
45
36
31
35
30
47
36
37
32
45
38
0.05
0.09
2.43
1.94
1.67
1.89
1.62
2.35
1.80
1.90
1.66
1.17
1.0
–
–
8
–
–
d
0.43
2.00
2.37
2.05
2.59
0.35
1.80
1.62
1.98
0.70
0.91
37
44
38
48
7
36
31
38
27
35
A(CVD,W1,C5) used
A(CVD,W10)
e
A(CVD,W10, C0.5) used 27
A(SSIE, W1)
A(SSIE, W1, C5)
B(CVD,W1)
32
30
28
26
B(CVD,W1,C5)
a
b
c
d
e
3+
Isolated Fe in tetrahedral and higher coordination.
Small oligomeric Fe O clusters.
x
y
Large Fe O particles.
2
3
Calcined at 823 K, spectrum not shown.
Spectrum in Fig. 6b.
three samples, but the absolute intensities decrease in the
order A(SSIE, W1) > A(CVD, W1, C0.5) > A(MR). The
signals at gꢀ ≈ 6 and gꢀ ≈ 4.3, and also the narrow one at
gꢀ ≈ 2 in A(MR), decrease with rising temperature as ex-
pected for paramagnetic behavior, whereby the intensity loss
of the gꢀ ≈ 4.3 signal is stronger, suggesting shorter relax-
ation times in comparison to the Fe3+ species reflected by
gꢀ ≈ 6. At T ꢁ 373 K those signals become narrower and
better resolved. This is attributed to the loss of water mole-
3.3. Influence of the mode of Fe incorporation
With the exception of the narrow signal at gꢀ ≈ 2,
the shape and behavior of the EPR signals in the low-
temperature range (90–270 K) are rather similar for the

390
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
cules from the pores which are assumed to be located in the
coordination sphere of the Fe3+ ions and give rise to a cer-
tain distribution of the zero field splitting parameters which
enhances the linewidth.
and Table 2 it is evident that sample A(MR) is dominated by
isolated Fe3+ sites with a certain amount of small oligomeric
FexOy moieties (band at 353 nm) also present. Those clus-
ters are dominating species in samples A(CVD, W1) and
A(SSIE, W1) while the latter contains also a considerable
amount of large Fe2O3 particles.
The broad signal at gꢀ ≈ 2 does not show Curie-like be-
havior. In the high-temperature range it increases and nar-
rows suddenly. This suggests that the samples contain anti-
ferromagnetically coupled FexOy species with a Neel tem-
perature of TN > 573 K for sample A(SSIE, W1) and TN >
373 K for samples A(CVD, W1, C0.5) and A(MR). Above
TN those species become paramagnetic and contribute to
the EPR signal. A strong line narrowing seen in particu-
lar in A(SSIE, W1) is due to spin–spin exchange interac-
tion which is most effective for well-ordered, large clusters.
Those species were earlier detected in sample A(SSIE) (i.e.,
without washing) by XRD, TPR, and XAFS [10]. It is not
unexpected that they remain also in the washed material.
Intensity increase and exchange narrowing are much less
pronounced in sample A(CVD,W1, C0.5); hence, the clus-
ters have a smaller size and a lower degree of order. In the
latter sample, the signal at gꢀ ≈ 2 is anisotropic. This sug-
gests that the signal is not due to a particular species with
defined geometry but to a superposition of FexOy species
with a certain size distribution. The results described con-
firm assumptions suggested in [10] to explain the striking
contradictions between EXAFS and Mössbauer results on
Fe-ZSM-5 prepared via the CVD route: disordered clusters
cannot be detected by methods that rely on wave interfer-
ence as XRD and EXAFS, but are seen by methods based
on magnetic interactions. The reason for the lack of order
in these clusters (as opposed to the extrazeolite aggregates
in A(SSIE, W1) is most likely an intrazeolite location; i.e.,
the zeolite structure may have been partly destroyed and
included during particle growth. This has been suggested
in [10] on the basis of XPS data and has been visualized
in [17] by scanning transmission electron micrographs.ꢀ
Note that the temperature dependence of the broad g ≈ 2
signal in sample A(MR) indicates clearly the presence of
FexOy clusters (Fig. 2c), although the lower intensity and
the larger linewidth suggest that they are much less abun-
dant and smaller than in samples A(SSIE, W1) and A(CVD,
W1, C0.5). As noted above, these species were not detected
by XAFS [20], illustrating the power of EPR spectroscopy
for such investigations. Even the parent H-ZSM-5 which
contains only 0.05% Fe shows a small gꢀ ≈ 2 line with non-
Curie behavior, indicating that it is not free of small FexOy
clusters. Moreover, signals at gꢀ ≈ 4.3 and 6 were not de-
tected in the parent H-ZSM-5. This shows clearly that the
respective Fe3+ species are introduced by the various Fe
loading procedures.
By taking into account both the EPR and UV/VIS results,
it can be concluded that each of the three Fe-ZSM-5 samples
contains at least two different Fe3+ single sites, probably
in tetrahedral and higher coordination as reflected by EPR
signals at gꢀ ≈ 4.3 and gꢀ ≈ 6 and by UV/VIS CT bands
below 300 nm. A third kind of isolated Fe3+ ions in less
distorted environment is present in A(MR), but it cannot be
excluded that it is present in the other samples as well, but is
masked by the EPR cluster signal at gꢀ ≈ 2. Besides the dif-
ferent isolated Fe3+ sites, FexOy aggregates are formed, the
amount and size of which increases considerably in the order
A(MR)
A(CVD, W1) < A(SSIE, W1). The data reveal
that the mechanochemical route is most effective in intro-
ducing preferably isolated Fe3+ sites into pore positions of
H-ZSM-5.
It should be noted that none of the characterization tech-
niques applied previously [10,20] was able to distinguish co-
existing different isolated Fe3+ species. Furthermore, XAFS
data of sample MR do not reflect any FexOy clusters, al-
though they are clearly detected by EPR and UV/VIS-DRS.
This illustrates the benefit of the two techniques for the char-
acterization of complex solid materials like Fe-ZSM-5.
3.4. Influence of heat treatment
The as-prepared states discussed above are not really
comparable because A(SSIE, W1) had been subjected to a
higher temperature than A(CVD, W1) during preparation.
Heat treatment; e.g., the calcination in air at 873 K, usually
performed before catalytic experiments [10,20] causes in-
deed significant structural changes as demonstrated in Fig. 4.
In this figure, EPR and UV/VIS spectra of sample A(CVD,
W1) are shown before calcination as well as after calcina-
tion in air at 873 K using heating rates of 5 and 0.5 K/min.
It is clearly seen from the EPR spectraꢀ(Fig. 4a) that signals
of isolated Fe3+ sites at gꢀ ≈ 6 and g ≈ 4.3 lose intensity
upon calcination while the signal at gꢀ ≈ 2 increases. This
suggests that initially isolated Fe sites aggregate to form
FexOy clusters. This effect seems to be slightly favored by
higher heating rates. In the corresponding UV/VIS spectra
(Fig. 4b), light absorption above 400 nm being characteris-
tic of extended Fe2O3-like clusters is somewhat higher after
calcination with 5 K/min than with 0.5 K/min (Fig. 4, Ta-
ble 2). Clustering of isolated Fe species upon calcination in
air has been observed for all FeZSM-5 samples including
A(SSIE, W1) (Table 2). It appears that FexOy cluster for-
mation is favored by moisture remaining in the zeolites after
washing.
Experimental UV/VIS spectra of the samples in the as-
prepared state (i.e., CVD and MR before calcination, while
SSIE is calcined during the preparation) are presented to-
gether with deconvoluted subbands in Fig. 3. The relative
percentage of the area of these subbands with respect to the
total spectral intensity is summarized in Table 2. From Fig. 3
EPR, on the other hand, reveals additional features of
the calcination process as shown in Fig. 5. In Figs. 5a and

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
391
Fig. 4. Structural changes of the iron phase in sample A(CVD, W1) during calcination as seen by EPR (a) and UV/VIS spectroscopy (b); spectra recorded
from hydrated samples at room temperature; before calcination (thick solid line), after calcination at 873 K with a heating rate of 0.5 K/min (dashed line), and
after calcination at 873 K with a heating rate of 5 K/min (thin solid line).
gꢀ ≈ 6 signal remains upon room-temperatureevacuation but
has lost much intensity to that at gꢀ ≈ 4.3 (Fig. 5b). When,
instead of 2 h evacuation at room temperature, the latter sam-
ple is again thermally treated in air at 773 K, very different
changes are induced (Fig. 5c):ꢀNow there is just a narrowing
of the signals at gꢀ ≈ 6 and g ≈ 4.3 and the effect is com-
pletely reversible upon exposure to the ambient atmosphere.
An explanation of this behavior may set out from the
assumption that [FeCl2]+ cations deposited on cation sites
during the CVD step [6,15] [Eq. (1)] are hydrolyzed during
the washing step. In this process, part of the iron species may
acquire a distorted coordination sphere with a coordination
number of 6 or 5 made up of water, charge-balancing OH
groups, and, possibly, oxygen of the zeolite wall, which is
reflected in the signal at gꢀ ≈ 6 [e.g., Eq. (2)]
H+ + FeCl3 → [FeCl2]+ + HCl,
(1)
(2)
ꢀ
ꢁ+
[FeCl2]+ +nH2O +mOz → Fe(H2O)n(Oz)m(OH)2
Fig. 5. Structural changes of the isolated Fe species in sample A(CVD, W1)
during calcination; EPR spectra recorded at room temperature; initial hy-
drated state (thick solid line), after 2 h evacuation at 293 K (thin solid line),
after 2 h calcination in air at 773 K (dotted line), and after reexposing to
the ambient atmosphere (dashed line); initial states: (a) as-prepared sample
A(CVD, W1); (b and c) calcined sample A(CVD, W1, C0.5) after long-term
storage at the ambient atmosphere.
+ 2HCl
(m + n = 3 or 4).
Upon room-temperature evacuation, the water ligands may
be reversibly removed, and the respective Fe species con-
tribute to the signal at gꢀ ≈ 4.3 which might arise from tetra-
hedral Fe species as frequently discussed in the literature
(Fig. 5a). After calcination, however, if contact with the hu-
mid air is avoided, the coordination of most Fe ions is much
different (Fig. 5c, dotted line). The difference is obviously
caused by the more severe effect of the thermal treatment as
opposed to room-temperature evacuation and consists pre-
sumably in a condensation of OH groups, e.g., the charge-
balancing OH groups on the Fe3+ ion. This process would
lead to a cation with an extraframework oxygen coordinated
to 4 or 5 framework oxygens (Eq. (3), with a higher proba-
bility for p = 4 due to the sterical conditions in the zeolite).
We believe that this species gives rise to the narrow signal at
gꢀ ≈ 6 after calcination (Fig. 5c), which is a typical feature
of this type of mononuclear Fe sites in the dehydrated state.
b, the effect of a room-temperature evacuation and subse-
quent contact with the ambient atmosphere is compared for
the uncalcined and the calcined samples A(CVD, W1) and
A(CVD, W1, C0.5), where the latter had been rehydrated
in ambient atmosphere between calcination and EPR experi-
ment for an extended period of time. Fig. 5c shows the effect
of a thermal treatment in air at 773 K (instead of room-
temperature evacuation) on the calcined and rehydrated sam-
ple A(CVD, W1, C0.5). It can be seen that evacuation of the
uncalcined A(CVD, W1) material leads to a complete dis-
appearance of the signal at gꢀ ≈ 6 while the one at gꢀ ≈ 4.3
increases strongly and becomes narrower (Fig. 5a). This shift
is largely reversed by contact with ambient atmosphere. For
the calcined but rehydrated sample A(CVD, W1, C0.5) the

392
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
There is, however, also a minority species in tetrahedral co-
row signal appears at gꢀ ≈ 2 above 373 K, indicating the
presence of extended FexOy clusters with marked antiferro-
magnetic coupling (Fig. 6c). In contrast, the increase of this
line in A(CVD, W10, C0.5) after catalysis is much less pro-
nounced, which suggests a weaker antiferromagnetic cou-
pling due to a smaller cluster size (Fig. 6d).
ordination (gꢀ ≈ 4.3).
ꢀ
ꢁ+
ꢀ
ꢁ+
Fe(H2O)n(Oz)m(OH)2 → (Oz)pFeO
(p = 4 or 5 for gꢀ ≈ 6, and p = 3 for gꢀ ≈ 4.3).
(3)
Upon contact with the humid ambient atmosphere, the
outer coordination sphere of the Fe species is influenced by
adsorbed water, which results in slightly different distortions
at different lattice positions and causes the broadening of
the signals, in particular that at gꢀ ≈ 6, which can be seen
in Fig. 5c. The reversal of the structural change upon dehy-
dration (probably rehydration of the Fe=O units), however,
appears to proceed much more slowly: This is suggested by
a comparison of the spectra after 2 h evacuation in Figs. 5a
and b. In contrast to the initial sample in Fig. 5a, the one in
Fig. 5b had been calcined but stored in ambientꢀatmosphere
for an extended period of time. The narrow g ≈ 6 signal
typical of the dehydrated structures is still left to an apprecia-
ble extent in the latter sample after 2 h evacuation (Fig. 5b);
i.e., prolonged storage in the ambient atmosphere led to only
incomplete rehydration of the dehydrated Fe site. Unfortu-
nately, it is difficult to predict on the basis of this information
which structure of the Fe cation will be present under reac-
tion conditions: while the higher temperatures should favor
dehydration, there is water present in the atmosphere in sig-
nificant amounts strongly depending on the water content of
the feed. This question will be subject to further in situ EPR
studies.
3.6. Influence of the SCR reaction
In Fig. 7, EPR and UV/VIS spectra of catalyst A(CVD,
W1, C0.5) before and after use in the SCR of NO with isobu-
tane (duration 1 working day) are compared in order to trace
structural changes that might occur in a precalcined catalyst
under reactionꢀconditions. In the EPR spectrum (Fig. 7a),
the signal at g ≈ 6 has almost disappeared while the sig-
nal at gꢀ ≈ 4.3 is virtually not effected. At the same time,
the cluster signal at gꢀ ≈ 2, has grown enormously. In the
corresponding UV/VIS spectrum (Fig. 7b), light absorption
above 400 nm reflecting extended Fe2O3-like clusters is only
slightly more pronounced after catalysis: The contribution of
the subbands above 400 nm to the total intensity increases
from 37 to 44% after the catalytic run (Table 2). This is al-
most exclusively at the expense of subbands between 300
and 400 nm while the contribution of the isolated sites re-
mains constant. Hence, during catalysis, large oxide parti-
cles are formed rather from small oligomeric moieties than
from isolated Fe sites.
This conclusion is also supported by the apparent contra-
diction between the extent of particle growth indicated by
the EPR and the UV/VIS spectra (Figs. 7a and b). While
the increase of the EPR cluster signal at gꢀ ≈ 2 by exposure
to the catalytic conditions is huge, the relative intensity of
the corresponding UV/VIS subbands above 300 nm changes
only slightly after reaction. Obviously, the strong increase
of the EPR signal at gꢀ ≈ 2 after reaction is mainly due to
intensified spin–spin exchange interactions between neigh-
boring Fe sites which is probably caused by lattice ordering
in larger FexOy clusters.
3.5. Influence of washing intensity
It has been reported in the literature that thorough wash-
ing is crucial for the formation of highly dispersed Fe struc-
tures [10,15,37]. In our study, the influence of the washing
step was checked by performing the washing procedure of
A(CVD) with a total amount of 1 L or 10 L water per 5 g cat-
alyst (A(CVD, W1) and A(CVD, W10), respectively.) The
UV/VIS spectra of these samples are compared in Fig 6a.
After washing with 10 L water, the light absorption above
400 nm is weaker in comparison to after the short wash-
ing procedure, which indicates that intense washing dimin-
ishes the amount of big FexOy clusters slightly (Table 1).
This has been concluded earlier from TPR and Mössbauer
measurements, although these had been performed with the
respective A(CVD, W1) and A(CVD, W10) samples after
calcination and use in the SCR reaction [10]. For compari-
son, the UV/VIS spectra of these (used) samples are given
in Fig. 6b. As shown above, thermal stress (calcination or
use in the SCR reaction, which is initiated by a thermal ac-
tivation/stabilization process) favors aggregation of isolated
Fe sites markedly. Even after use in catalysis, however, the
better dispersion of the iron species in the intensely washed
material can be traced by a decreased absorption above
400 nm (solid line). This is also confirmed by temperature-
dependent EPR measurements (Figs. 6c and d). In the spec-
tra of A(CVD, W1, C0.5) after catalysis, an intense nar-
The strong attenuation of the gꢀ ≈ 6 line suggests that
Fe species reflected by this signal, are modified in the cat-
alytic reaction while those related to the gꢀ ≈ 4.3 line remain
unchanged. The UV/VIS data (Table 2), on the other hand,
imply that this modification is not caused by cluster forma-
tion or reduction but just by a change in the Fe coordination
geometry. In our recent in situ EPR studies of the direct N2O
decomposition and the SCR of N2O by CO over different
Fe-MFI catalysts we have observed that the gꢀ ≈ 6 EPR sig-
nal changes upon contact with feed components while the
line at gꢀ ≈ 4.3 remains virtually unchanged [38]. Similar
effects were found during in situ EPR investigations of sam-
ples A(MR) and A(CVD, W1, C0.5) during the SCR of NO
by i-butane [39]. Other authors also observed that the gꢀ ≈ 6
signal in Fe-ZSM5 vanishes after use of the material as a cat-
alyst for different reactions, which was attributed to the high
activity of these particular Fe3+ sites [2,40,41]. Kubánek et
al. derived a linear correlation between the gꢀ ≈ 6 signal in-

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
393
Fig. 6. Influence of the washing intensity on the Fe dispersion in Fe-ZSM-5. UV/VIS spectra of (a) as-prepared hydrated samples A(CVD, W1) (dashed line)
and A(CVD, W10) (solid line) and (b) calcined hydrated samples A(CVD, W1, C0.5) (dashed line) and A(CVD, W10, C0.5) (solid line) after use in the SCR
reaction; EPR spectra of the used hydrated samples A(CVD, W1, C0.5) (c) and A(CVD, W10, C0.5) (d). Samples were recalcined after catalysis in air at
823 K.
Fig. 7. Changes of the Fe species after use in the SCR with isobutane. EPR (a) and UV/VIS spectra (b) of hydrated sample A(CVD, W1, C0.5) measured at
room temperature before (thick solid line) and after use in the SCR of NO (broken line). Samples were recalcined after catalysis in air at 823 K.
tensity and the activity of Fe-ZSM-5 in the conversion of
benzene to phenol by N2O while no change of intensity was
observed for the gꢀ ≈ 4.3 signal [42]. It was concluded that
Fe3+ species reflected by the latter signal are mainly located
in extraframework Al–O clusters that had been formed by a
steaming pretreatment and, therefore, do not participate in
the catalytic reaction.
Fe-ZSM-5 catalysts used in the present work have not
been pretreated by steaming. Moreover, no indication for
a partial lattice damage upon use in the SCR reaction was
found that could give rise to dealumination and inclusion of
Fe in an extraframework Al–O phase. However, it is pꢀossible
that the Fe species responsible for the unchanged g ≈ 4.3
signal may be unable to extend their coordination sphere
by adsorbing reactant molecules. It seems not unlikely that
those Fe3+ species could be located in the γ -sites of the
MFI structure. It was found that the latter represent sites of
strongest distortion [43].ꢀThis would agree pretty well with
the fact that a signal at g ≈ 4.3 arises, too, from Fe3+ sites
in the strongest possible distortion. Although Fe3+ species
in γ sites should be accessible by reactant molecules from
the main channel [43], their confinement within the rather
narrow pocket-like g site could prevent the necessary coor-
dination of reactant molecules from the gas phase.
As suggested by EPR and UV/VIS results discussed
above (Fig. 7, Table 2), large oxide particles seem to be
preferably formed from small oligomeric moieties during
use in the SCR reaction while the amount of isolated Fe
sites remains virtually unchanged. Interestingly, no deactiva-
tion occurs on the time scale of the experiments performed
despite those significant structural changes. The fact that
the amount of small oligonuclear clusters decreases in fa-

394
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
Fig. 8. Influence of the Si/Al ratio and the lattice perfection on the structure of Fe species in Fe-ZSM-5. EPR and UV/VIS spectra (measured from hydrated
samples at room temperature) of materials prepared with zeolites ZSM-5(A) and ZSM-5(B), (a) as-prepared samples B(CVD, W1) (dashed line) and A(CVD,
W1) (solid line); (b) calcined samples B(CVD, W1, C5) (dashed line) and A(CVD, W1, C5) (solid line).
vor of large oxide particles while isolated Fe sites persist
during reaction may indicate that the latter are the active
Fe sites of the SCR reaction. However, the changes between
the amounts of oligomers and large aggregates (Table 2) are
too small to safely exclude a contribution of the clustered
phases. On the other hand, since EPR reflects a strong order-
ing tendency in the clustered phase and it is unlikely that the
catalytic activity of clusters will be improved by increasing
structural perfection, the structural changes observed on the
whole support the view that isolated Fe sites play a deciding
role in the SCR reaction as proposed, for instance, in [10,19,
20].
Interestingly, this trend is not retained for the calcined
samples (Fig. 8b). Calcination with a heating rate of 5 K/min
leads to a slightly less pronounced formation of large Fe2O3-
like aggregates in sample B(CVD, W1, C5) compared to
A(CVD, W1, C5) as evident from both EPR and UV/VIS
spectra. This may be explained by the presence of silanol
nests in the sample, which may serve as additional nuclei
for aggregation as has been suggested earlier on the basis of
FTIR, Mössbauer, and EXAFS data [10]. In well-structured
HZSM-5 matrices FexOy clusters being formed upon cal-
cination have been shown to migrate toward the external
crystal surface where they can grow further in size [17]. In
contrast, silanol nests present in highly defective matrices
could keep the clusters inside the crystal, thus preveꢀnting
their further growth. Remarkably, the EPR signal at g ≈ 6
is extremely weak or completely absent in sample B(CVD,
W1, C5); instead, the signal at gꢀ ≈ 4.3 is more pronounced.
3.7. Influence of the Al content and defect density of the
parent ZSM-5 matrix
Fe-ZSM-5 was also prepared via CVD of FeCl3 into
the ZSM-5 matrix B, which has a lower Si/Al ratio than
Matrix A (≈ 40 vs ≈ 14). This, in turn, leads to a lower
number of H+ sites available for exchange by Fe ions,
and a large amount of internal silanol groups, most likely
silanol nests [10]. As expected, the total Fe content in sam-
ple B(CVD, W1) is lower than in sample A(CVD, W1), but
the Fe/Al ratio of 1.13 shows that sample B(CVD, W1) is
even more overexchanged than the latter (Fe/Al ≈ 0.9). It is
therefore in agreement with expectations that the EPR sig-
nals for isolated Fe sites at low magnetic field are less intense
for the uncalcined sample B(CVD, W1) than for A(CVD,
W1) while the cluster signal at gꢀ ≈ 2 is larger (Fig. 8a).
A slightly higher percentage of extended Fe2O3-like clus-
ters in sample B(CVD, W1) is also evident from UV/VIS
data (Fig. 8a, Table 2).
3.8. Active sites for SCR catalysis
The characterization studies confirm the structural com-
plexity of the iron species obtained by CVD of FeCl3 into H-
ZSM-5 and related methods and support the skepticism to-
ward quantitative conclusions from EXAFS spectra of such
materials, as has been expressed earlier in [10,11]. This com-
plexity is a serious obstacle for the identification of the active
sites for selective NO reduction. However, by comparing
trends in the SCR activity data and the structural data ob-
tained in this study, some suggestions can be derived even
from the work with these highly nonideal materials.
An effort to find correlations between the concentration
of Fe species as analyzed by UV/VIS (Table 2) and cat-
alytic properties may embark from the very similar behavior

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
395
of some catalysts with both the isobutane and the ammo-
nia reductant (reference group A(SSIE, W1, C5), A(CVD,
W1, C5), and A(CVD, W10, C0.5)). The comparison of the
species distribution in these materials (Table 2) shows that
the amounts of all sites vary to a certain extent so that a
correlation is not possible. This is not unexpected because
the species distinguished by UV/VIS in fact represent sev-
eral entities, e.g., at least 3 isolated sites, one of which
(gꢀ ≈ 4.3 in EPR) appears to be unable to extend its coor-
dination sphere by adsorbing molecules from the gas phase,
oligomers of different nuclearity (starting with dimers), and
particles of different size. The comparison with the activity
data of A(MR, C5) shows, however, that the participation
of isolated sites in the catalytic reactions is highly likely.
This sample, which achieves the highest normalized reaction
rates in both reactions (Table 1), and with isobutane even
the highest NO conversions in a broad temperature range,
does not contain particles and only a very small amount of
oligomers while the quantity of isolated sites is somewhat
lower than in the reference-group catalysts. While particles
can be therefore safely rejected as active sites, the extremely
low quantity of oligomers (more than an order of mag-
nitude less than in the reference-group catalysts) strongly
discourages an assignment of the catalytic activity exclu-
sively to oligomers. The data imply rather that mononuclear
Fe species participate in SCR catalysis, probably together
with oligomeric entities. A similar suggestion has been made
above for isobutane-SCR on the basis of the observation that
significant ordering of the clustered species occurs during
catalysis (Fig. 7) without concomitant changes in the cat-
alytic behavior.
For NH3-SCR, the participation of several types of sites
is nicely illustrated by the behavior of the catalyst prepared
with the silica-rich matrix H-ZSM-5(B) – B(CVD, W1, C5).
While this material fails completely in isobutane-SCR, it is
ranked between A(MR, C5) and the reference-group cat-
alysts in NH3-SCR (Fig. 1), where it exhibits almost the
same normalized reaction rate as the reference-group cata-
lysts (Table 1). The considerably higher activity as compared
with A(MR, C5) (Fig. 1) is achieved with an only slightly
larger smaller amount of isolated entities (0.68 vs 0.41 wt%,
cf. Table 2); hence aggregated species will contribute to ac-
tivity. On the other hand, the latter are hardly present in
A(MR, C5); hence they cannot be the single type of active
site for NH3-SCR. It appears therefore that all iron accessi-
ble may be part of active sites for NH3-SCR. In addition, it
should be noted that B(CVD, W1, C5) has only a very low
acidity as has been found in FTIR [10] and ammonia-TPD
studies. This indicates that acidity is not of importance for
NH3-SCR with Fe-ZSM-5, at least in the presence of ac-
tive sites of the clustered type. The reason for the failure of
B(CVD, W1, C5) with isobutane cannot be conclusively ex-
plained. It may be due to the absence of a particular Fe site
that is summarized but not distinguished by UV/VIS (e.g.,
a particular mononuclear site) or to the absence of sufficient
acidity. The role of acidity in isobutane -SCR with Fe-ZSM-
5 will be discussed elsewhere.
In NH3-SCR, NO conversion drops at high temperatures
for all catalysts except A(MR, C5). The decreased NO con-
version is accompanied by an almost constant NH3 conver-
sion of ca. 90% (not shown); i.e. part of the reductant is
oxidized at higher temperatures to N2 [4,43]. With A(MR,
C5), NH3 conversion equals NO conversion within the limits
of experimental accuracy up to 873 K (not shown). From this
it may be concluded that the ammonia oxidation is catalyzed
by the clustered species (oligomers, particles), although the
details of the high-temperature behavior (differences be-
tween A(SSIE, W1, C5), A(CVD, W1, C5), A(CVD, W1,
C0.5)) are not yet understood. A similar discussion is pos-
sible for isobutane-SCR: Here, the catalyst with a negligi-
ble amount of clustered entities achieves high NO conver-
sions still at 700–800 K where those with large amounts of
oligomers and particles are much less selective for NO re-
duction. The competition of nonselective oxidation starts at
a lower temperature for the isobutane than for the ammo-
nia reductant, which may reflect a different propensity of the
reductants to become oxidized over Fe sites. From this, an
improved catalyst for both SCR processes should contain
increased amounts of mononuclear species with as low as
possible quantities of clustered entities. Studies of catalysts
prepared with this intention are currently under way.
4. Conclusions
Fe-ZSM-5 catalysts prepared by different techniques
from H-ZSM-5 (solid-state ion exchange, chemical vapor
deposition of FeCl3 with variations of the subsequent wash-
ing and calcination steps, mechanochemical treatment with
FeCl3), which had been previously characterized by XRD,
EXAFS, Mössbauer spectroscopy, TPR, FTIR, and XPS,
have been additionally analyzed by EPR and UV/VIS diffuse
reflectance spectroscopy (DRS)—ex situ after preparation,
after calcination and after use in the selective catalytic reduc-
tion of NO as well as in situ during calcination. It has been
found that combined EPR and UV/VIS-DRS measurements
are powerful tools for elucidating the nature of coexisting Fe
species formed in the Fe-ZSM-5 materials investigated. The
new structural data have been discussed with regard to the
catalytic properties of the materials in the selective catalytic
reduction of NO by isobutane or NH3.
In the materials prepared, three different isolated sites
have been identified by EPR spectroscopy, which have been
assigned toꢀ Fe3+ in tetrahedral (gꢀ ≈ 4.3) and higher coor-
dination (g ≈ 6), and to Fe3+ in a highly symmetric envi-
ronment (gꢀ ≈ 2). The latter has been observed only in the
sample with the lowest Fe content, A(MR) at low tempera-
tures. In samples with higher Fe contents an intense signal
of aggregated iron oxide species appears at the same g value
of 2. UV/VIS-DRS differentiates between isolated Fe3+
sites, the coordination of which cannot be distinguished,

396
M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
small (oligomeric) Fe oxide clusters, and large Fe oxide ag-
gregates. The degree of aggregation and the order of the ag-
gregates can also be studied by temperature-dependent EPR
measurements where it is reflected in increasing intensity of
the cluster signal due to breakdown of the antiferromagnetic
coupling. Based on these assignments, the following conclu-
sions could be derived:
References
[1] G.I. Panov, G.A. Sheveleva, A.S. Kharitonov, V.N. Romannikov, L.A.
Vostrikova, Appl. Catal. A 82 (1992) 31.
[2] A. Ribera, I.W.C.E. Arends, S. de Vries, J. Perez-Ramirez, R. Sheldon,
J. Catal. 195 (2000) 287.
[3] J. Perez-Ramirez, F. Kapteijn, G. Mul, X. Xu, J.A. Moulijn, Catal.
Today 76 (2002) 55.
[4] A.-Z. Ma, W. Grünert, Chem. Commun. (1999) 71.
[5] X. Feng, W.K. Hall, Catal. Lett. 41 (1996) 45.
[6] H.-Y. Chen, W.M.H. Sachtler, Catal. Today 42 (1998) 73.
[7] R.Q. Long, R.T. Yang, Catal. Lett. 74 (2001) 201.
[8] G. Berlier, G. Spoto, S. Bordiga, G. Ricchiardi, P. Fisicaro, A. Zec-
china, I. Rossetti, E. Selli, L. Forni, E. Giamello, C. Lamberti, J. Ca-
tal. 208 (2002) 64.
[9] A.A. Battiston, J.H. Bitter, D.C. Koningsberger, J. Catal. 218 (2003)
163.
[10] F. Heinrich, C. Schmidt, E. Löffler, M. Menzel, W. Grünert, J. Ca-
tal. 212 (2002) 157.
[11] E.J.M. Hensen, Q. Zhu, M.M.R.M. Hendrix, A.R. Overweg, P.J. Kooy-
man, M.V. Sychev, R.A. van Santen, J. Catal. 221 (2004) 560.
[12] R.Q. Long, R.T. Yang, J. Am. Chem. Soc. 121 (1999) 5595.
[13] Q. Sun, Z.-X. Gao, H.-Y. Chen, W.M.H. Sachtler, J. Catal. 201 (2001)
88.
[14] A.A. Battiston, J.H. Bitter, D.C. Koningsberger, Catal. Lett. 66 (2000)
75.
[15] P. Marturano, L. Drozdova, A. Kogelbauer, R. Prins, J. Catal. 192
(2000) 236.
[16] J. Jia, Q. Sun, B. Wen, L.X. Chen, W.M.H. Sachtler, Catal. Lett. 82
(2002) 7.
[17] A.A. Battiston, J.H. Bitter, F.M.F. de Groot, A.R. Overweg, O. Ste-
phan, J.A. van Bokhoven, P.J. Kooyman, C. van der Spek, G. Vanko,
D.C. Koningsberger, J. Catal. 213 (2003) 251.
All preparations (including the mechanochemical route)
lead to the coexistence of different iron species. In all cases,
isolated Fe sites are found, which are predominant after the
mechanochemical route (which leads to low Fe content), but
minority species with the other preparations. In the CVD
preparation, clustered species occur already after the wash-
ing step, but their quantity and size increase strongly during
calcination, with low gradients of temperature rise resulting
in higher Fe dispersions. Intense washing of the material af-
ter the CVD step also favors higher Fe dispersion but does
not prevent clustering. With a matrix of low density of Brøn-
sted sites but high defect density, a low Fe dispersion before
calcination was transformed into a highly disperse (but still
largely clustered) Fe phase, probably due to the internal de-
fects providing additional aggregation nuclei. Calcination
leads also to a significant change in the coordination sphere
of isolated Fe ions, which might be caused by a condensa-
tion of two charge-+balancing OH groups to an O2− ligand
(Fe(OH)2+ → FeO ). The rehydration of the calcined state
is slow at room temperature.
Comparison of the distribution of species detected by
UV/VIS spectroscopy with activity trends in NH3-SCR sug-
gests that different entities (isolated sites, oligomers, prob-
ably also the surface of particles) participate in this reac-
tion. The acidity of the zeolite is of minor importance with
catalysts that contain clustered entities. For isobutane-SCR,
the participation of isolated sites is suggested by the re-
markable performance of a catalyst that is almost void of
clustered entities. A participation of isolated sites is also
inferred from the observation that during the use of the
catalysts in the SCR reaction on a timescale where no de-
activation was noted, oligomeric clusters tend to aggregate
and achieve a more perfect order while the amount of iso-
latedꢀsites remains constant. Among the isolated sites, one
(at g ≈ 4.3) appears to be insensitive to influences from
the gas phase. Possible candidates for the active site are the
signals at gꢀ ≈ 6 and gꢀ ≈ 2 (isolated species), where the
latter is difficult to observe. The observation that nonselec-
tive oxidation of the reductant (isobutane or NH3) starts at
higher temperatures with a catalyst of low content of clus-
tered species suggests that aggregates favor this undesired
reaction.
[18] R. Joyner, M. Stockenhuber, J. Phys. Chem. B 103 (1999) 5963.
[19] Z. Sobalik, A. Vondrova, Z. Tvaruskova, B. Wichterlova, Catal. Today
75 (2002) 347.
[20] F. Heinrich, C. Schmidt, E. Löffler, W. Grünert, Catal. Commun. 2
(2000) 317.
[21] H.-Y. Chen, El-M. El-Malki, X. Wang, R.A. van Santen, W.M.H.
Sachtler, J. Mol. Catal. A: Chem. 162 (2000) 159.
[22] A. Brückner, R. Lück, W. Wieker, B. Fahlke, H. Mehner, Zeolites 12
(1992) 380.
[23] D. Goldfarb, M. Bernardo, K.G. Strohmaier, D.E.W. Vaughan, H. Tho-
mann, J. Am. Chem. Soc. 116 (1994) 6344.
[24] A. Brückner, U. Lohse, H. Mehner, Micropor. Mesopor. Mater. 20
(1998) 207.
[25] A.V. Kucherov, C.N. Montreuil, T.N. Kucherova, M. Shelef, Catal.
Lett. 56 (1998) 173.
[26] A. Brückner, G.-U. Wolf, M. Meisel, R. Stösser, H. Mehner, F. Ma-
junke, M. Baerns, J. Catal. 154 (1995) 11.
[27] S. Bordiga, R. Buzzoni, F. Geobaldo, C. Lamberti, E. Giamello,
A. Zecchina, G. Leofanti, G. Petrini, G. Tozzola, G. Vlaic, J. Catal. 158
(1996) 486.
[28] J.H. Bitter, A.A. Battiston, S. van Donk, K.P. de Jong, D.C. Konings-
berger, Micropor. Mesopor. Mater. 64 (2003) 175.
[29] A. Brückner, B. Kubias, B. Lücke, Catal. Today 32 (1996) 215.
[30] P. Wenquin, Q. Shilun, K. Zhiyun, P. Shaoyi, Stud. Surf. Sci. Catal. 49
(1989) 281.
[31] A.F. Ojo, J. Dwyer, R.V. Parish, Stud. Surf. Sci. Catal. 49 (1989)
227.
[32] R. Aasa, J. Chem. Phys. 52 (1983) 3919.
[33] B.G. Fox, J.D. Lipscomb, in: C.C. Reddy, G.A. Hamilton, K.M.
Madyastha (Eds.), Biological Oxidation Systems, vol. 1, Academic
Press, San Diego, CA, 1990, p. 367.
Acknowledgment
[34] H.H. Tippins, Phys. Rev. B 1 (1970) 126.
[35] G. Lehmann, Z. Phys. Chem. Neue Folge 72 (1970) 279.
[36] X. Gao, I.E. Wachs, J. Phys. Chem. B 104 (2000) 1261.
We thank the Deutsche Forschungsgemeinschaft (DFG)
for financial support (Grants Br 1380/7-1 and Gr 1447/7-1).

M.S. Kumar et al. / Journal of Catalysis 227 (2004) 384–397
397
[37] P. Marturano, L. Drozdova, D. Pirngruber, A. Kogelbauer, R. Prins,
Phys. Chem. Chem. Phys. 3 (2001) 5585.
[40] A.V. Kucherov, C.N. Montreuil, T.N. Kucherova, M. Shelef, Catal.
Lett. 56 (1998) 173.
[38] J. Pérez-Ramirez, M. Santhosh Kumar, A. Brückner, J. Catal. 223
(2004) 13.
[41] A.M. Volodin, V.I. Sobolev, G.M. Zhidomirov, Kinet. Catal. 39 (1998)
775.
[42] P. Kubánek, B. Wichterlová, Z. Sobalík, J. Catal. 211 (2002) 109.
[43] J. De˘decˇek, D. Kaucký, B. Wichterlová, Micropor. Mesopor. Mater.
35–36 (2000) 483.
[39] M. Santhosh Kumar, M. Schwidder, W. Grünert, U. Bentrup, A. Brück-
ner, in: 18th North American Catalysis Society Meeting, June 1–6,
2003, Cancun, Mexico, 2003, p. 4, Book of abstracts.