ϭ
Controlling collective properties of supported metal
nanoparticles: towards stable catalysts
Gonzalo Prieto
1
, Jovana Zeþeviü
1
, Heiner Friedrich
2
, Krijn P. de Jong
1,
* and Petra E. de Jongh
1,
*
1
Inorganic Chemistry and Catalysis, Debye Institute for Nanomaterials Science, Utrecht University, The
Netherlands
2
Laboratory of Materials and Interface Chemistry, Eindhoven University of Technology, Eindhoven, The
Netherlands
*
Corresponding authors: k.p.deJong@uu.nl and p.e.deJongh@uu.nl
Supported metal nanoparticles play a pivotal role in areas such as nanoelectronics, energy
storage/conversion
[1]
and as catalysts for the sustainable production of fuels and chemicals
[2,3,4]
. However, the
tendency of nanoparticles to grow into larger crystallites is an impediment for stable performance
[5,6]
.
Exemplary, loss of active surface area by metal particle growth is a major cause of deactivation for
supported catalysts
[7]
. In specific cases particle growth might be mitigated by tuning the properties of
individual nanoparticles, such as size
[8]
, composition
[9]
and interaction with the support
[10]
. Here we present
an alternative strategy based on control over collective properties, revealing the dramatic impact of the 3D
nanospatial distribution of metal particles on catalyst stability. We employ silica-supported copper
nanoparticles as catalysts for methanol synthesis as a showcase. Achieving near-maximum interparticle
spacings, as accessed quantitatively by electron tomography, slows down deactivation up to an order of
magnitude compared to a catalyst with non-uniform nanoparticle distribution, or a reference Cu/ZnO/Al
2
O
3
catalyst. Our approach paves the way towards rational design of practically relevant catalysts and other
nanomaterials with enhanced stability and functionality, for applications such as in catalysis, sensors, gas
storage, batteries, and solar fuels production.
Metal nanoparticle growth can proceed via migration and coalescence of particles (sintering) or via
transport of monoatomic or molecular species between individual particles (Ostwald ripening)
[7]
. Strategies to
mitigate particle growth comprise alloying with a higher-melting point metal
[9]
and increasing the metal-support
interaction energy by using specific oxides as carrier
[10]
. However, these approaches are not generally applicable
since they restrict the catalyst chemical composition and therefore function. Recently, the encapsulation of
individual colloidal nanoparticles in porous inorganic shells has received much attention
[11,12]
. Albeit conceptually
elegant, bottom-up approaches at the single-nanoparticle level face challenges for large scale production and usage.
We present an alternative approach, using nanoparticle assembly tools to tune the stability-relevant collective
properties of supported metal particles, i.e. their spatial distribution at the nanoscale.
Metal nanoparticles dispersed on a porous carrier are widely used as solid catalysts
[13]
. Irregular spatial
distributions and ultra-short interparticle distances are quite common for technical catalysts. For instance,
Ϯ
clustering of metal nanoparticles in high-density assemblies is known to occur for Ni-Mo/Al
2
O
3
and Co-Pt/Al
2
O
3
catalysts industrially employed in processes such as the sulphur-removal from hydrocarbon feedstocks and the
Fischer-Tropsch synthesis of fuels from synthesis gas, respectively
[13,14]
.
Also, commercial methanol synthesis
catalysts typically consist of Cu particles (ca. 5-10 nm) only spaced by smaller ZnO crystallites
[15]
.
Nanoparticles
spatial distributions might have large significance for catalyst stability, given that metal particle growth is a
relevant deactivation mechanism for commercial catalysts. Though on a micrometer-scale metal distributions can
be controlled to some extent (for instance “egg-shell” or “egg-yolk” distributions), the 3D spatial distribution of
metal particles on a nanometers scale is until now typically an uncontrollable outcome of the preparation process.
In contrast, our work shows control over size and location of active metal species while employing industrially
relevant preparation tools, in this case impregnation of porous carriers with inexpensive metal precursors such as
nitrates
[16,17,18]
.
Quantitative information on the 3D nanospatial distribution of the metal particles is indispensable if we
want to understand supported nanoparticles stability, and progress from 2D flat model substrates towards more
realistic 3D support morphologies. This information is provided by electron tomography (ET), which has emerged
as a major tool in material science
[19,20,21]
. ET allows derivation of metal particle size distributions and nanospatial
locations. In this work we additionally employ pore-specific analysis, which relates particle locations to the local
support pore morphology. We combine these tools to demonstrate exceptional stability by control over the
nanospatial distribution of supported metal particles, using Cu-Zn/SiO
2
catalysts for methanol synthesis as a case in
point.
The industrial production of methanol from synthesis gas (CO/CO
2
/H
2
) amounts worldwide to more than
35·10
6
t/year and uses Cu/ZnO/Al
2
O
3
as a catalyst. Growth of the Cu crystallites is the main deactivation pathway
under standard plant conditions. In this study we employ as support material ordered mesoporous silica (SBA-
15
[22]
) rather than industrial Al
2
O
3
to facilitate quantitative image analysis. Also samples based on industrial SiO
2
-
gel supports with 3D interconnected pore networks were investigated to validate the wider significance of our
results. The CuZn/SiO
2
catalysts were prepared by impregnation using an aqueous solution of metal nitrates.
Building on recent mechanistic insight
[17]
, for the first time we succeeded in preparing exclusively <6 nm Cu
nanoparticles at relatively high metal loadings. In particular, effective water removal at low temperatures was
essential to avoid the large metal agglomerates and bimodal size distributions previously observed
[17]
(see
Supplementary Methods). The dried impregnate was heated to 723 K (referred to as calcination hereafter) under
either N
2
or 2%NO/N
2
flow. Samples are labelled as CuZn/x(y), where x=S (SBA-15) or Sgel (SiO
2
-gel) and y= N
2
or NO according to the calcination atmosphere. A Cu/ZnO/Al
2
O
3
reference catalyst was prepared by co-
precipitation route representative for commercial catalysts
[23]
. This reference catalyst showed similar copper-
weight-based initial catalytic activity, validating our approach to use an ordered silica model support.
Table 1 gives an overview of the structural properties of the catalysts. Mean Cu
0
crystallite sizes were 2.4-
9.7 nm after H
2
-reduction, whereas crystalline Zn compounds were not detected. Zn species are essential in
promoting the catalytic Cu particles to achieve optimum methanol productivity
[24]
. Close contact between Cu and
ϯ
Zn (oxide) is prerequisite for this synergism
[25]
. X-ray absorption spectroscopy showed the high dispersion of Zn
species on the surface of the silica support (Supplementary Fig. S1). Energy-dispersive X-ray spectroscopy (EDX)
evidenced that the required intimacy between Zn and Cu on the nanoscale is achieved regardless of the calcination
procedure (Supplementary Fig. S2). Hence, hereafter the discussion will focus on the Cu particles. The CuZn/SBA-
15 catalysts exposed similar Cu
0
surface areas (104-118 m
2
/g
Cu
), approximately double that of the Cu/ZnO/Al
2
O
3
reference sample, albeit at four times lower Cu content. Thus, bulk characterization techniques point to little
difference between the SiO
2
-supported catalysts obtained by calcination in either N
2
or 2%NO/N
2
flow. However,
these characterization methods lack spatial resolution. Scanning-transmission electron microscopy (STEM) gave a
first indication of different spatial distribution of the active Cu phase (Supplementary Fig. S2). Nonetheless,
conventional (S)TEM implies averaging information over the entire thickness of the support particle, e.g. here >20
superimposed mesopores.
Table 1 | Structural and catalytic properties of the studied catalysts.
Catalyst S
B.E.T
[a]
[m
2
g
SiO2
-1
]
PV
meso
[b]
[cm
3
g
SiO2
-1
]
Cu
[c]
[wt-%]
S
Cu
[d]
[m
2
g
Cu
-1
]
d(Cu)
XRD
[e]
[nm]
MeOH
yield
[f]
[mol g
Cu
-1
h
-1
]
MeOH
select.
[g]
[C-%]
k
D,2
[h]
[10
-3
h
-1
]
SBA-15 775 0.79 --- --- --- --- --- ---
CuZn/S(NO) 657 0.68 12.2 118 3.5 0.11 98.3 4.4
CuZn/S(N
2
) 623 0.71 11.5 104 2.4 0.09 97.8 0.9
SiO
2
-gel 354 1.00 --- --- --- --- --- ---
CuZn/Sgel(NO)
CuZn/Sgel(N
2
)
318 0.81 11.8 --- 5.9 0.10 98.3 20.3
286 0.78 10.7 --- 9.7 0.11 98.5 5.9
Cu/ZnO/Al
2
O
3
--- --- 51.4 59 5.1 0.13 99.6 8.9
[a]
Specific B.E.T. surface area.
[b]
Mesopore volume.
[c]
Copper loading.
[d]
Specific Cu surface area for the as-reduced catalyst, as
determined by N
2
O reactive frontal chromatography.
[e]
Mean Cu
0
crystallite size for the as-reduced catalyst, as determined by X-ray
diffraction-monitored in situ H
2
-reduction.
[f]
Initial methanol yield; reaction conditions: T=533 K, P=40 bar, synthesis gas feedstock
Ar:CO
2
:CO:H
2
=10:7:23:60 (vol), initial (CO+CO
2
) conversion of 15-20%.
[g]
Steady (time-on-stream>100 h) selectivity to methanol
(dimethylether is the only significant side-product)
[h]
Second-order deactivation rate constant obtained after fitting the deactivation profile
with a second-order deactivation law: da/dt=-k
D,2
·a
2
, where a denotes the normalized methanol productivity.
We used ET combined with image analysis to provide precise 3D information on the size, shape and
location of the individual Cu nanoparticles. The size and position of all particles in a tomogram was obtained by
automatic segmentation (Supplementary Information, section 5.1), but more importantly also pore-specific manual
segmentation was applied as illustrated in Figure 1. This analysis allowed acquisition of the in-pore surface-to-
surface interparticle distances, which are presumably crucial for growth, irrespective of whether caused by sintering
or by Ostwald ripening.
ϰ
Figure 1 | How statistically relevant, pore-specific information on the size and 3D nanospatial distribution of metal
particles is derived from electron tomograms. a, Orthogonal cross-sections through the 3D reconstructed tomogram of a
representative particle of a nanostructured catalyst. The SiO
2
pore walls are grey and the metal nanoparticles appear as black
spots. b, 3D-rendered volume obtained by image segmentation showing the hexagonal arrangement of the catalyst mesopores
as well as details on the internal corrugation of the silica pore walls. One of the reconstructed mesopores has been partially
sectioned to visualize the individually-segmented Cu nanoparticles. c, 3D-view of the isolated Cu nanoparticles extracted from
the volume indicated with a yellow frame in panel b. The approximation of the metal particles by volume-equivalent spheres
centred at their center of gravity and the derivation of surface-to-surface interparticle distances is schematically depicted. For
validation of the statistical relevance of our analysis results see sections 5.1 and 5.2 in the Supplementary Information.
[Figure 1: 17 x 7 cm (double column)]
Figure 2 summarizes the electron tomography results for the as-prepared CuZn/SBA-15 catalysts after H
2
-
reduction, giving cross-sections through the reconstructed 3D volumes (a-b, e-f), Cu particle size (c,g) and
neighbor-distance distributions (d,h). Full electron tomograms are given as Supplementary Videos 1 and 2. For
CuZn/S(NO) (panels a-d) a fraction of the pores contained high metal loadings, while other pores hosted virtually
no metal particles. A surface-averaged Cu particle size of 5.7±2.0 nm was derived from the corresponding size
histogram. The uneven distribution of the catalytic particles in the pore system resulted in distance distributions
centred at 2.7 and 7.9 nm for the nearest and second-nearest neighbors, respectively. Such irregular spatial
distribution and short interparticle distances are quite common for technical catalysts.
ϱ
Figure 2 | Quantitative electron tomography results for CuZn/SBA-15 catalysts after H
2
-reduction. The figure includes
cross-sections through the 3D-reconstructed tomograms (a-b and e-f), surface-weighted Cu nanoparticle size histograms (c,g)
and surface-to-surface nearest-neighbors distance histograms (d,h) for the Cu particles in the CuZn/SBA-15 catalysts calcined
under a flow of 2%NO/N
2
(a-d) or a flow of N
2
(e-h), and subsequently reduced in 20%H
2
/N
2
. The values for the average and
standard deviation are included in the histograms. Cu particles hosted in the same silica pore were considered to generate the
interparticle distance histograms.
[Figure 2: 17 x 8.5 cm (double column)]
Recent insight into ex-nitrate catalyst synthesis suggested strategies to control the spatial distribution of the
active Cu species
[17]
. Starting point was the distribution of the metal nitrate precursors within the support porosity
after impregnation and drying (Supplementary Fig. S3). Calcination in the presence of NO promoted hydrolysis
forming Cu
2
(OH)
3
NO
3
which has a low surface mobility. Minimized metal redistribution led to the typical
patchwise distribution discussed above. However, by efficient vacuum-drying at room temperature followed by
calcination in N
2
flow we prevented premature hydrolysis, enabling now full exploitation of Cu redispersion via a
mobile anhydrous Cu(NO
3
)
2
intermediate over the SiO
2
surface. Despite the similarities in average structural
properties accessed by bulk characterization techniques, ET revealed a large impact of the calcination procedure on
the nanospatial distribution of the Cu nanoparticles.
The N
2
-calcined catalyst (panels e-h) comprised Cu particles with a narrow size distribution (4.0±1.4 nm)
which, in marked contrast to CuZn/S(NO), were evenly distributed throughout the entire pore system. This
homogeneous distribution is further illustrated by the broad neighbor-distance distributions (Fig. 2h). The average
distances of 13.9 and 33.2 nm to the nearest and second-nearest neighbors, respectively, were 4-5 fold higher than
for CuZn/S(NO). Remarkably, they were very close to the theoretical maximum neighbor distances (14.9 and 31.1
nm) for 4 nm-sized Cu particles at the given Cu loading. Hence, CuZn/S(N
2
) featured a distribution which seems
ideal for a growth-resistant catalyst, i.e. equally sized (minimum driving force for ripening) and maximally spaced
metal nanoparticles. We focused on catalysts supported on SBA-15 silica as their well defined pore structure
facilitates visualization and quantitative analysis of the results, although correspondingly, Cu nanoparticles
