Nucleation and Growth Mechanism of Semicoductors

Journal of Colloid and Interface Science 317 (2008) 351–374
www.elsevier.com/locate/jcis
Feature article
Nanocluster nucleation and growth kinetic and mechanistic studies:

A review emphasizing transition-metal nanoclusters ✩
Eric E. Finney, Richard G. Finke ∗
Department of Chemistry, Colorado State University, Fort Collins, CO 80523, USA
Received 8 December 2006; accepted 18 May 2007
Available online 24 August 2007
Abstract
A review of the literature of kinetic and mechanistic studies of transition-metal nanocluster nucleation and growth is presented; the focus
is on nucleation processes. A brief survey of nucleation theory is given first, with an emphasis on classical nucleation theory, as this is the
logical starting point of transition-metal nanocluster nucleation and growth studies. The main experimental methods for following nanocluster
formation are examined next—dynamic light scattering, UV–visible spectroscopy, electron microscopy, and X-ray spectroscopies—with special
attention paid to their strengths and weaknesses. Several specific examples of transition-metal nanocluster formation are then given, beginning
with LaMer’s classic sulfur sol system and including the Finke–Watzky mechanism of slow continuous nucleation A → B followed by fast
autocatalytic surface growth A + B → 2B. Finally, brief overviews of semiconductor nanoparticle preparations, solid-state nucleation studies—
emanating from Avrami’s work—and protein agglomeration mechanistic studies are also provided, as these processes are relevant, conceptually
and in a general sense, to the field of transition-metal nanocluster nucleation and growth mechanisms.
© 2007 Elsevier Inc. All rights reserved.
Keywords: Nanoclusters; Nucleation; Growth; Kinetics and mechanism
1. Introduction Avrami’s work and the subsequent work that it inspired), and
protein aggregation, topics which are examined briefly in Ap-
Nucleation and growth—the appearance of a new phase out pendices A, B, and C. An experimental-based understanding
of an old phase—is a ubiquitous phenomenon in nature [1]. The of nucleation in these processes is limited, the reason being
process occurs in everything from cloud formation to volcano illustrated by the statement that “the understanding of crystal-
eruptions to nanocluster formation reactions. In fact, it has been lization, especially the nucleation process, has been very much
suggested that “it should not be a surprise if it proves that even limited mainly due to a lack of quantitative experimental data”
the Big Bang was a nucleation phenomenon” [1]. The majority [5]. The nucleation, and subsequent growth, of soluble nano-
of the theoretical work with nucleation theory (i.e., the devel- clusters in solution has received relatively little prior attention,
opment and modification of classical nucleation theory, vide this contribution being the first review of the subject.
infra) has been done for the condensation of liquid from the Recently, however, the nucleation and growth of metal nano-
vapor phase [2]. This work has been extended to the nucleation clusters in solution has been the subject of increasing study.
of solid from the melt (i.e., freezing) [3], as well as precipita- It is well known that at the nanometer scale, the optical, elec-
tion and crystallization of solids from solution [4]. Nucleation tronic, and catalytic properties of transition-metal nanoclusters
and growth has received a great deal of study in the areas of are highly sensitive to their size and shape [6]. Therefore,
semiconductor particles, solid-state nucleation (most notably one of the main goals of nanocluster science is the ability
to prepare nanoclusters that have very narrow size distribu-
✩
tions (so-called “near-monodisperse” nanoclusters of ±15%
Based on the literature review chapter of E.E. Finney’s Ph.D. thesis at Col-
orado State University, 2008. [7]), thereby allowing for greater uniformity of nanocluster
* Corresponding author. properties. A fundamental understanding of the mechanism of
E-mail address: rfinke@lamar.colostate.edu (R.G. Finke). nanocluster formation should allow greater control over nano-
0021-9797/$ – see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcis.2007.05.092
352 E.E. Finney, R.G. Finke / Journal of Colloid and Interface Science 317 (2008) 351–374
cluster size, shape, and composition (e.g., of multimetallic general expression is given in Eq. (2) [11]:
“nano-onions” [8]), resulting in the ability to tune the above-
x
mentioned properties simply by varying the reaction conditions. G = −nkB T ln sat + σ bn2/3 , (2)
This review will focus on the nucleation of soluble nano- x
clusters in solution, but brief mention of the vapor and pre- where n is the number of atoms in the cluster, kB is the Boltz-
cipitation systems will be given first, as they provide the basis mann constant, T is the temperature, σ is the surface tension
for the nucleation theory applied to other areas of science, in- of the cluster, b is a geometric factor, and the quantity x/x sat is
cluding transition-metal nanoclusters. Reviews of nucleation, called the supersaturation of the solution, where x is the mole
especially of classical nucleation theory and its derivatives, fraction of the species present in solution, and x sat is the mole
are plentiful [9,10], and excellent texts giving detailed descrip- fraction of the species in a saturated solution. Alternatively, the
tions of general nucleation theory exist [1,11]. Therefore, this supersaturation can be expressed in terms of concentration, and
review will only briefly summarize the dominant theories cur- becomes C/Ceq , where Ceq is the solubility (the equilibrium
rently in practice, with appropriate references for the interested concentration) of the species. When the expression is used for
reader; the focus of this review is the application of nucleation the nucleation of droplets from vapor, the supersaturation is
to transition-metal nanoparticles in solution. expressed in terms of pressure, p/peq . The free energy of for-
Although nucleation in general has been extensively stud- mation of nuclei is at the heart of CNT, it being noted that “the
ied, the high levels of accuracy and precision that are enjoyed exact determination of G (for the critical nucleus) seems to
by other fields of science remain elusive in the study of nu- be the real problem of nucleation theory” [18].
cleation. A general shortcoming of nucleation theory is that When comparisons are made to experimental work, the nu-
much of the theory comes from fitting experimental data and cleation rate (denoted as either I or J ) is generally the pa-
adjusting theoretical parameters accordingly. As a result the rameter that is measured. This parameter is also expressed in
theory in general has little predictive power [12]. Therefore, different ways; one of these is given in Eq. (3) [19]:
a main goal of nucleation theory is to improve the accuracy of
the theory to better match, and then accurately predict, exper- 16π γ3
I = Γ exp − , (3)
imental observations [13]. The large gap between a theoretical 3kB T (ρs |μ|)2
(mathematical) understanding of nucleation and an experimen-
where ρs is the number density of the solid, γ is again the
tal (physical) understanding is a gap that is only recently being
solid–liquid interface free energy density, μ is the change in
bridged by physicists and physical chemists. As Oxtoby has
chemical
potential between the solid and the liquid phases, and
observed, “nucleation theory is one of the few areas of sci-
Γ =√ Zρl fc+ . In this last expression, Z is the Zeldovich fac-
ence in which agreement of predicted and measured rates to
tor, |μ|/6πkB T nc , where nc is the number of particles in
within several orders of magnitude is considered a major suc-
the critical nucleus and fc+ is the attachment rate of particles
cess” [9]. Zukoski notes that errors of tens of orders of mag- 2/3
nitude are not unprecedented [14]! Closer agreement between to the cluster, which is given by fc+ = 24DλS2nc ; here, DS is a
theory and experiment is, therefore, a goal of nucleation sci- self-diffusion coefficient and λ is a diffusion distance. An ex-
ence. pression for the nucleation rate is given as Eqs. (4) and (5). In
Eq. (4), B is pre-exponential factor which is a collection of the
2. Classical nucleation theory terms described above, and G∗ is the activation energy for
nucleation.
What is now known as classical nucleation theory (CNT)
G∗
was pioneered by Volmer [15,16] and Becker and Döring [17]. J = B exp − , (4)
CNT uses the Gibbs capillary approximation, which treats the RT
nucleus as if it were a bulk phase—which, rigorously, it is not. 16πr 3 Vm2
G∗ = . (5)
The free energy of nucleus formation is expressed in different 3|μ|2
ways, but every expression contains two competing terms: one
The dependence of the free energy of a cluster as a function
negative, for the favorable formation of bonds in the nucleus,
of the radius of the cluster, based on classical nucleation theory,
and one positive, for the unfavorable creation of an interface
is shown qualitatively in Fig. 1 [11].
between the nucleus in its new phase and the medium of the old
Because the cluster free energy has negative (bond mak-
phase. In its simplest form and in terms of the cluster radius, the
ing) and positive (surface energy) terms, the free energy has
expression for the cluster free energy is that given in Eq. (1) [9]:
a maximum at some radius, called the critical radius, r ∗ . Dif-
4 ferentiating Eq. (1) with respect to r and substituting back into
G = − πr 3 |Gv | + 4πr 2 γ . (1)
3 Eq. (1) gives an expression for the maximum free energy, G∗ ,
This equation assumes a spherical nucleus, as this shape min- Eq. (6), where Gv = −(RT /Vm ) ln S, where Vm is the molar
imizes the surface area of the nucleus. Here, r is the radius of volume of the relevant species and S is the degree of supersat-
the cluster, Gv is the difference in bulk free energy per vol- uration. Note that Eq. (6) has the same general form as Eq. (5).
ume between the old and new phases, and γ is the surface free
16πr 3
energy per unit area. The cluster free energy can also be ex- G∗ = . (6)
pressed in terms of the number of atoms in the cluster; one such 3|Gv |2
E.E. Finney, R.G. Finke / Journal of Colloid and Interface Science 317 (2008) 351–374 353
critical temperature, while other data are fit better when actual
experimental dγ /dT data are used [20].
Computationally, Grànàsy developed diffuse interface the-
ory to predict nucleation rates of nonane from the vapor, and
then compared his calculated values to experiment along with
calculations using CNT [22]. Diffuse interface theory contains
two assumptions that are somewhat relaxed compared to CNT:
(i) that only the very center of the nucleus, but not the entire nu-
cleus, has the properties of the bulk, and (ii) that the thickness
of the liquid–vapor interface is independent of the nucleus size.
These modifications gave results that were closer to experiment
than CNT, further showing how the very basic assumptions of
CNT fail to account for experimental observations. The diffuse
Fig. 1. Dependence of the free energy, G, of a cluster on the radius of the interface theory was also compared to CNT in a computational
cluster, r. The curve has a maximum free energy G∗ at the critical nucleus, study of the temperature dependence of the nucleation rate of
which has a radius of r ∗ .
gold particles during freezing, using clusters composed of var-
ious numbers of gold atoms [23]. In this case, similar results
were obtained—CNT overestimated the nucleation rate at high
This maximum of the curve, where dG/dr = 0, gives the
temperatures, while diffuse interface theory fit the observed nu-
radius of the critical nucleus of the cluster. For clusters smaller
cleation rate.
than r ∗ , growth is unfavorable and the cluster dissolves. For
The validity of CNT has been examined experimentally,
clusters larger than r ∗ , growth is favored. Alternatively, the crit-
mostly in systems of liquid droplet formation from the vapor.
ical nucleus can be expressed in terms of the number of atoms
For example, El-Shall et al. used vapor-phase nucleation of do-
contained within it, n∗ , by differentiating Eq. (2) and finding the
decane, hexadecane, and octadecane to check the accuracy of
maximum. For clusters containing less than n∗ atoms, growth
CNT [24,25]. They found that the theory breaks down at low
is unfavorable and the cluster dissolves. For clusters containing
temperatures (T ∼300 K). At these temperatures, CNT pre-
more than n∗ atoms, growth is favored. The critical nucleus is
dicts a higher level of supersaturation needed for nucleation
therefore the activated complex along the reaction coordinate of
than was found experimentally, and it underestimates the num-
nucleation [9].
ber of atoms in the critical nucleus, n∗ , for their system [24]. In
The greatest flaw of CNT is that it treats nuclei as bulk
accordance with this data, the authors used the scaling model
material having macroscopic properties. The key assumptions
of Hale [21] and added extra terms to the theory to compen-
involved here are that (i) the nucleus behaves as the bulk, and
sate for the temperature dependence of the nucleation rate [24].
(ii) the surface free energy of the cluster is the same as that
By adding a term for the increased surface entropy of the clus-
of an infinite planar surface [9–11]. Both of these assumptions
ter, the theory became much better at predicting the nucleation
become unreasonable when considering a nucleus of only sev-
rate over the temperature range studied, 280–350 K. The clas-
eral (or even several hundred) atoms or molecules; the center of
sic work of Strey et al. in 1994 [26] also tested the validity
the cluster is far from the thermodynamic limit, and the sharp
of CNT in the condensation of 1-butanol from the vapor us-
curvature of the cluster significantly increases the free energy
ing a “nucleation pulse chamber” to measure nucleation rates.
of its surface. In the context of Eq. (3) above, accurate defin-
These authors obtained a result similar to that of El-Shall et al.,
itions and values of the constants λ and γ are unavailable for
specifically that CNT accurately describes nucleation of liquid
small nuclei vs bulk material. As stated by Martínez et al. in the
droplets at high temperature, but is less accurate at lower tem-
case of liquid–vapor nucleation, “clearly the use of bulk liquid
peratures (240 K).
properties to describe a process involving small metallic states
One of the difficulties of using many of the various nucle-
is problematic” [20]. Obviously, the same can be said for solid
ation theories is that they contain the γ term for the interfacial
particles.
energy of the cluster. The value of this parameter is not known
In 1986, Hale reported a scaled nucleation theory (SNT)
for small nuclei [19]. An attempt to solve this problem has been
[21], which introduced critical temperature and pressure, Tc and
made by Sugimoto et al., who formulated an expression for the
pc , as well as a new parameter Ω for the excess surface entropy.
solid–liquid interfacial energy [27,28], and then used their for-
This new parameter effectively corrects for the supersaturation
mulation to calculate the value of γ for silver halide particles
and therefore for the nucleation rate and nucleation free energy.
[29]. A brief review of studies attempting to obtain the interfa-
The value of Ω is dependent on density, which must be approxi-
cial energy is given in Ref. [29]. Their result is shown in Eq. (7):
mated for non-bulk systems. Unfortunately, a consistent method
of approximating Ω is not available. Different researchers have γ = −N σ λkB T ln x(∞) , (7)
obtained values of Ω using different methods [20]. The fitting
of experimental data to SNT shows that no one method for ap- where N σ is the surface density of surface monomers (a “mono-
proximating Ω is universal—some of the experimental data are mer” in the case of AgX clusters studied here is defined as a
fit better when the surface tension is approximated using the single AgX unit, where X = Cl, Br, I), λ is the ratio of open
sites on a surface monomer to that of a free monomer, kB is 3. Nanocluster formation in solution

the usual Boltzmann constant, T is the temperature, and x(∞)
is the solubility of the bulk material given by the mole fraction 3.1. Methods used to study nanocluster formation
of monomers in the liquid phase. The authors were able to use
textbook values for the AgX crystals, such as the lattice para- Early measurements of the formation of nanoclusters were
meter and molar volume, to calculate γ for the clusters. They visual. In the work of Nielson in the 1960s, solutions were
also showed theoretically that γ is independent of the size of mixed and the time it took for a precipitate to be observed was
the cluster, and then verified this experimentally [29]. measured [36,37]. Since that time, a handful of spectroscopic
In a separate report, the surface tension of poly(methyl techniques have become common for measuring the kinetics of
methacrylate) spheres was estimated using laser scanning con- nanocluster formation. Some of those methods are mentioned
focal microscopy [30]. The results were in “reasonable agree- briefly here. A main problem with nearly all methods is that
ment” [30] with computational results. In addition, the shapes they tend to measure larger clusters—that is, growth—not the
of the nuclei were observed to be roughly elliptical, rather than desired nucleation.
spherical.
3.1.1. Dynamic light scattering (DLS)
Nucleus shape considerations Dynamic light scattering is used to obtain information about
particle sizes in solution, as well as interactions between macro-
It has been found using computational studies that under- molecules such as proteins [38]. In chemistry, DLS is a com-
neath a face-centered cubic (fcc) crystal nucleus is a layer mon technique for measuring the sizes of polymer clusters [39],
with body-centered cubic (bcc) symmetry [31]. This finding has but is less common for the measurement of nanoclusters, the
been investigated as an extension of Ostwald’s “step rule” [32] dominant technique being ex situ transmission electron mi-
of solid-state chemistry, which states that the structure of the croscopy (vide infra). Particle size from fitting DLS data re-
nucleus that is first formed does not have to be the most ther- quires knowledge of the diffusion coefficient of the clusters and
modynamically stable, but is the one that most resembles the viscosity of the solvent. DLS is advantageous in that it is non-
parent solid phase. Density functional theory (DFT) calcula- invasive and is performed on a solution sample, so that the clus-
tions performed by Oxtoby [33] corroborated this finding, with ter solution requires little manipulation and can be recovered
the conclusion that metastable states are crucial in the forma- after analysis. In the nanocluster literature, DLS measurements
tion of nuclei. This result is perhaps not surprising given that are often compared to measurements from other techniques,
the critical nucleus itself is a metastable state [9,11]. A “Bain most commonly transmission electron microscopy; the agree-
transformation” is one where a bcc critical nucleus yields a fcc ment is generally close [40], at least in cases where no sample
crystal (or, also, the reverse process) [33]. damage has occurred from electron microscopy.
A similar result was found by Frenkel et al. in the formation
of bimetallic Pt–Ru particles [34]. Fits to EXAFS data gave 3.1.2. UV–visible spectroscopy
evidence of an fcc structure for the particles, while bulk Pt– The most popular method that is used to study the kinet-
Ru alloys have a hexagonal close packed (hcp) structure. The ics of nanocluster formation in solution as well as the sizes of
final nanocluster structure was determined to also have an fcc nanoclusters is UV–visible spectroscopy. Although this is a use-
structure. Yet another example is given in the diffuse interface ful method for some systems, it requires that the nanoclusters
theory study of freezing gold particles mentioned above [23], under investigation absorb in the UV–visible region, and this
in which the nuclei assume an hcp structure as opposed to the tends to restrict such studies to the metals with distinct plas-
more stable fcc structure. Some transformation from hcp to fcc mon absorbances in the visible region, namely Ag, Au, and
was observed for some of the gold nuclei, but the calculated Cu [41]. This tends to exclude nanoclusters of many interest-
barrier for the transformation was large. ing metals, including Co, Ni, Re, Ir, and Pt, that primarily tail
Cacciuto and Frenkel have noted that the principle of Ost- off into the UV region [41]. Therefore, other methods are neces-
wald’s step rule is not taken into account by CNT [35]. They sary to study the kinetics of nanocluster nucleation and growth.
found that CNT cannot account for stress that exists within Alternatively, UV–visible spectroscopy can be used to follow
crystal nuclei arising from the formation of a surface; CNT as- the loss of the nanocluster precursor, provided that the precur-
sumes that the nuclei are incompressible, and this assumption, sor absorbs in the UV–visible region. An example of such use is
the authors claim, is unnecessary [35]. The stress experienced the study of Pt0n nanoclusters from the reduction of [PtII Cl4 ]2−
by the nuclei can influence the structure of the nuclei; therefore, by Henglein et al. (vide infra) [42].
understanding the stress that nuclei experience can enable pre- One major problem with studying nanocluster formation us-
dictions about the structure of the nuclei in conjunction with the ing UV–visible spectroscopy is that unless the formation can be
step rule. Using theoretical studies, the authors found that un- studied in situ (i.e., without significant manipulation of the sam-
der certain conditions small crystal nuclei are less dense (i.e., ple), there is no way to be certain that the observed behavior of
experience less stress) than large ones. This fact, which likely the nanoclusters is the same as their actual behavior under the
has an effect on nucleation rates and the critical nucleus, is not specific reaction conditions. There have been attempts to obtain
taken into account by CNT. UV–visible spectra of nanoclusters in situ; one of these meth-
ods, used in the formation of CdSe nanocrystals, is given in make these techniques among the most attractive for study-
Appendix A. ing nanoclusters [55]. Frenkel and coworkers have pioneered
Once an in situ UV–visible method for measuring nucleation the use of these X-ray spectroscopies to study the formation of
kinetics is achieved, another issue arises—namely, that there are transition-metal nanoclusters [34,55–57]. Using advanced mod-
often numerous other species in the reaction medium that may eling techniques, they were able to study the shape of bimetal-
absorb in the UV–visible region and interfere with the nano- lic Pt–Ru nanoclusters described above [34]. In addition, they
cluster spectra. Among these are metal ions, solvent, reductant, found that early in the formation of the Pt–Ru nanoclusters, Pt
and added stabilizers. This problem has also been addressed is found preferentially at the core of the nuclei; as the nano-
[43,44]; the results are presented in Section 4.6. clusters grow, a core–shell inversion takes place, after which
Pt is found mainly on the surface of the clusters [56]. Along
3.1.3. Electron microscopy with determining the structure of the nuclei vs the structure
Another common method used in nanocluster formation of the final nanocluster, the shapes of carbon-supported clus-
studies is ex situ electron microscopy. The most popular mi- ters were determined to be hemispherical; this was found to
croscopy used in the nanocluster literature is transmission elec- be the case for both Pt–Ru bimetallic nanoclusters and carbon-
tron microscopy (TEM). TEM is often used to characterize the supported Pt nanoclusters [55]. The wealth of structural infor-
growth of nanoclusters, as nuclei are typically too small to mation that is possible from X-ray spectroscopies makes this
be seen, non-high resolution, “routine” TEM typically being technique very promising in the characterization of transition-
limited to 1 nm despite stated resolution limits typically of metal nanocluster nucleation if it can be done in situ. Appro-
± 0.2 nm. Recently, however, the method of high-angle annu- priate EXAFS, XANES, and other studies of Ir(0)n and other
lar dark-field (HAADF) imaging with scanning TEM (STEM) transition-metal nanocluster formation are being pursued col-
has become available [45], so-called “Z-contrast” microscopy. laboratively between the Nuzzo, Frenkel, and Finke laborato-
This method, which has increased contrast in heavier atom sam- ries [58].
ples, has been shown to achieve atomic resolution—for exam- To summarize, the best characterization techniques for
ple, in one study, germanium and erbium atoms were visualized studying the nucleation and growth of transition-metal nano-
on a SiC support [45]. In general, TEM offers several advan- clusters are those that can be performed in situ/operando, al-
tages, including powerful visual images and direct measure- lowing for direct observation of the nucleation and growth
ment of nanocluster size. However, because TEM is a solid- processes. Electron microscopy fails at this, despite its rel-
state (i.e., ex situ) technique, the nanoclusters must be taken out ative ease and the visually striking images that it provides.
of solution and dried before they can be analyzed, during which UV–visible and X-ray spectroscopies, however, require less
further growth and agglomeration of the nanoclusters in the sample manipulation and may be performed in situ. Using
increasingly concentrated solution (i.e., before achieving dry- these techniques in combination, while keeping in mind the
ness) can take place. In addition, some monometallic and other, advantages and limitations of each, is necessary to gain a better
especially first- or second-row transition-metal nanocluster pre- picture of what is happening during nanocluster nucleation and
cursor complexes yield nanoclusters in the TEM beam [46,47]. growth.
Hence, TEM alone cannot tell if nanoclusters are nucleating in
solution or being produced in the TEM beam. The TEM beam 3.2. LaMer’s experimental cluster nucleation studies
may also change the nanoparticles under observation, another
major disadvantage, one that has been studied in detail [48]. The most famous and widely cited example of the applica-
A TEM textbook notes that “in the end, you can damage vir- tion of nucleation theory to cluster formation is the work done
tually anything you put into the TEM” [49]. For these reasons, by LaMer in the 1950s [59,60]. LaMer used CNT to describe
TEM alone is not a reliable method to study the nucleation of the formation of sulfur sols from the decomposition of sodium
nanoclusters. thiosulfate in hydrochloric acid, Scheme 1. The model he pre-
Another notable method for visualizing nanoclusters is scan- sented, Fig. 2, shows the concentration of nucleating species
ning tunneling microscopy (STM) [50–52]. Again, however, (sulfur in this case) vs time. The essence of the mechanism,
STM studies are confined to nanoclusters examined under ultra- as illustrated by Fig. 2, is as follows: the concentration of ele-
high vacuum conditions on a support [53]. mental sulfur builds up slowly until some critical concentration
is reached, or more specifically a critical supersaturation level,
3.1.4. X-ray spectroscopies C/Ceq where Ceq is the solubility of sulfur in the solution (vide
Two potentially powerful methods for directly observing supra), stage I in Fig. 2. At this point, “self-nucleation” (i.e.,
the nucleation process utilize X-ray spectroscopy—X-ray ab- homogeneous nucleation) occurs at a rate that has been called
sorption near-edge spectroscopy (XANES) and extended X-ray “effectively infinite” [59], stage II; this is the origin of the idea
absorption fine structure spectroscopy (EXAFS) [54]. These of “burst” nucleation. The burst of nucleation immediately low-
methods are valuable because they give information about the
identity of atoms in the nanoclusters, as well as information
about the nearest neighbors of the atoms, allowing atomic-level
characterization. The recent, major advances in X-ray optics
as well as the analysis of multiple-shell scattering currently Scheme 1. The system studied by LaMer.
Fig. 2. The LaMer mechanism of nucleation of sulfur. The (theoretical) curve

shows sulfur concentration as a function of time. Adapted from Ref. [59].
Fig. 3. Nucleation curve for the reduction of HAuCl4 . Reprinted from Ref. [65]
ers the supersaturation level of monomers in the solution; as by permission of The Royal Society of Chemistry.
a result, nucleation essentially stops at this time. Growth then
occurs by diffusion of sulfur atoms throughout the solution,
stage III.
LaMer’s mechanism has been cited many times in the stud-
ies of nucleation of various clusters; for example, a Web of
Science search reveals that LaMer’s 1950 paper has been cited Scheme 2. Oxidation of citric acid to acetonedicarboxylic acid.
>380 times. While it may be a viable mechanism for the spe-
cific system of sulfur sol formation (and others, vide infra), it cleation [65]. In their study, they reduced chloroauric acid
is not expected to be the mechanism by which transition-metal in the presence of deprotonated citrate, C6 H5 O3− 7 , as a sta-
nanoclusters nucleate and grow from dilute solutions. Despite bilizer at different temperatures and different Au and citrate
this, LaMer’s pioneering mechanistic work described above concentrations. They measured nucleation rates using electron
(Scheme 1) has been applied to nanocluster preparations [61]. microscopy (a newly developed technique at that time). They
This has led the opinion that LaMer’s work has been “overcit- found that particle formation followed kinetics consisting of an
ed” [61]; Finke and Watzky have pointed out that this statement induction period, followed by a fast increase in the number of
is unfair to LaMer and coworkers, given that LaMer’s mecha- particles, followed by a linear increase, followed by a rapid de-
nism was the only one available for almost 50 years [62]! crease in the rate, Fig. 3. In their “organizer theory,” various
Studies of AgX clusters by Sugimoto et al. building off both species in the reaction solution bind to the gold ions to form
his previous work [27,28] (vide supra) and LaMer’s mecha- “copolymers” of gold ion and the “organizer.” The “organizer”
nism were directed at predicting the number density of the AgX in this case was proposed to be the citrate ion, which would be-
clusters and the time at which nucleation ends and growth become oxidized to acetonedicarboxylic acid upon reduction of
gins [63]. Using the LaMer mechanism and the condition of the gold ions, Scheme 2.
continuously added monomers, Sugimoto et al. developed a Further evidence presented for this process was that when
quantitative theory of nucleation, and then verified it experi- acetonedicarboxylic acid was added to chloroauric acid instead
mentally via the formation of AgCl [63] and AgBr [64] clusters. of citrate, the nucleation curve did not show an induction pe-
Essentially, these authors showed that the LaMer mechanism riod. Therefore, the induction period was proposed to be the
can be applied to the formation of AgX clusters from a su- time that is necessary for the citrate ion to become oxidized to
persaturated solution. It is interesting to note that the authors acetonedicarboxylic acid, which rapidly reduces the gold ions
suggest a “sort of Ostwald ripening” [63] during the nucleation to Au0 . Considering the Finke–Watzky mechanism discussed in
process, in which larger nuclei grow at the expense of smaller a moment, it seems likely that the induction period is actually
nuclei. This is effectively another way of saying that the nucle- due to the formation of critical Aun nuclei.
ation process is an equilibrium one until the critical nucleus is The “organizer” model, while obscure in its chemical details
attained, with nuclei dissolving and their monomers going to in comparison to LaMer’s straightforward, 2-step mechanism,
other nuclei. is experimentally well-supported and was a step toward a more
relevant mechanism for the formation of transition-metal nano-
4. Experimental studies of transition-metal nanocluster clusters. However, Turkevich’s mechanism uses only words to
nucleation and growth describe the nanocluster nucleation and growth process. Re-
stated, the “organizer” model lacks precise chemical equations
4.1. Turkevich’s organizer theory of nucleation that sum to the observed stoichiometry and which define the
rate constants of the nucleation and growth steps, keys to a more
Shortly after LaMer’s work, in 1951, Turkevich and cowork- detailed mechanistic understanding of the nanocluster forma-
ers put forth what they called the “organizer” model for nu- tion process.
Scheme 3. The Finke–Watzky 2-step mechanism for transition-metal nano-

cluster nucleation (rate constant k1 ) and growth (rate constant k2 ). A represents
the nanocluster precursor, and B represents the surface of the growing cluster.
Fig. 4. Typical cyclohexene loss vs time curve and curve-fit for the reduction
of [Bu4 N]5 Na3 [(1,5-COD)Ir·P2 W15 Nb3 O62 ] under H2 . The fit is obtained
using non-linear least-squares methods to the analytic equation, Eq. (8), for
A → B, A + B → 2B autocatalytic mechanism, Scheme 3. Of the >700 ki-
netic experiments done to date, the fits are generally good to excellent. When
Scheme 4. The cyclohexene hydrogenation reporter reaction used to measure some deviations vs the experimental data are seen, as in the above, deliberately
the kinetics of nanocluster formation. chosen example of one of the poorer fits, these can often be quantitatively ac-
counted for by a 3- or 4-step mechanism, vide infra.
4.2. The Finke–Watzky 2-step, slow, continuous nucleation

and then fast autocatalytic growth, mechanism
The next step in the study of transition-metal nanocluster

formation did not come until 1997, when Finke and Watzky de-
scribed a mechanism for the reduction of transition-metal salts
under H2 that consists of slow, continuous (not burst) nucle-
ation followed by fast, autocatalytic (not diffusion-controlled)
surface growth, Scheme 3 [62].
In the initial studies, A represents the nanocluster precur-
sor complex [Bu4 N]5 Na3 [(1,5-COD)Ir·P2 W15 Nb3 O62 ], and B
represents the growing surface of the Ir0n nanocluster.1 The ki-
netics of nanocluster nucleation and growth are measured using
the reporter reaction of cyclohexene hydrogenation, Scheme 4;
what is actually measured is the loss of hydrogen. In this way,
in situ/operando kinetic measurement of the nanocluster nucle-
ation and growth is achieved, yielding well-defined rate con-
stants k1 and k2 for nucleation and growth, respectively. The
rate constants are determined by fitting the hydrogen loss data Scheme 5. As discussed in the main text, other intermediates besides Ir0 , in
(converted for convenience as shown in Fig. 4 to the equivalent particular Ir–H (iridium hydrides), are possible if not probable. The individual
cyclohexene loss by the known 1:1 stoichiometry) to the ana- steps shown are postulated, and have not been directly detected.
lytic rate equation, Eq. (8), derived from the 2-step mechanism
in Scheme 3.
(but less precise) gas–liquid chromatography (GLC) measure-
k1
k2 + [A]0 ment of cyclooctane evolution from reduction under H2 of
[A]t = k1
. (8)
1+ k2 [A]0 exp(k1 + k2 [A]0 )t the 1,5-cyclooctadiene present in the nanocluster precursor,
[Bu4 N]5 Na3 [(1,5-COD)Ir·P2 W15 Nb3 O62 ], demonstrates that
It should be noted that although the rate constants are de- the rate constants calculated from both methods are the same
termined using the indirect cyclohexene hydrogenation re-
within experimental error [8]. Following the reduction by 1 H
porter reaction, control experiments using the more direct
NMR also shown that the cyclohexene reporter reaction method
provides accurate rate constants [66].
1 It must be remembered that the steps in Scheme 3 are composite, pseudo-
It is important to point out at this stage that the composite,
elementary steps [62], and therefore there are kinetically hidden mechanistic
steps in the overall, multistep nanocluster formation mechanism for example as pseudo-elementary steps [62], A → B and A + B → 2B, are
shown in Scheme 5. really something more like the following, Scheme 5.
Fig. 5. The kinetic data from Turkevich’s system of reduction of HAuCl4 , fit to Scheme 6. Illustration of the 4-step mechanism for transition-metal nano-
the 2-step mechanism in Scheme 4 [62]. The data was converted to give a curve cluster nucleation, growth, and then bimolecular plus autocatalytic agglomera-
that represents the loss of HAuCl4 , and then was fit to the analytic equation (8) tion [66,75].
using Microcal Origin: k1 = 3(1) × 10−4 h−1 ; k2 = 30(2) M−1 h−1 .
The problem of integral calculus2 of such steps is treated analogous autocatalytic mechanism. This fact, and Fig. 5, have
in a simplest (Ockham’s Razor), average way to start (as good not been previous published.
mechanism must) via the A → B, A + B → 2B 2-step mecha- An important synthetic insight gained from the autocatalytic
nism. The rate constants k1 and k2 are, therefore and necessar- mechanism in Scheme 4 is the need to avoid heterogeneous
ily, averages. An average “xgrowth ” factor has been introduced nucleation when forming nanoclusters [71]. Heterogeneous nu-
[62] to account for the changing surface area of the nanocluster, cleation of metal on solid surfaces is often energetically more
and a more continuous way to treat that change is forthcoming favorable (lower G‡ ) than homogeneous nucleation [26]. In
[67]; see also [68]. Importantly, numerical integration simula- addition, since heterogeneous nucleation is often highly vari-
tions done elsewhere [69] of the first ca. 50 of the above steps able, reproducible nanocluster formation is not possible in the
show that (i) a sigmoidal curve just like that in Fig. 4 is ob- presence of heterogeneous nucleation. It has been found that the
tained, and (ii) that resultant curve is fit extremely well by the nucleation rate constant k1 varies by 103 when heterogeneous
Finke–Watzky, 2-step kinetic model (see Fig. 7 presented else- nucleation occurs whereas k1 varies by 101.2 when homo-
where [69]). geneous nucleation occurs [71]. Heterogeneous nucleation can
The separation of nucleation and growth in time, necessary lead to the formation of thin metal films, which also have been
for narrow size distributions, is achieved in this system as seen shown to form by the autocatalytic A → B, A + B → 2B mech-
visually by the induction period followed by fast cyclohexene anism [71]. One can question whether or not nucleation is ever
loss in the overall sigmoidal curve in Fig. 4. The ratio of growth truly homogeneous in such systems, as typically the amount of
and nucleation rates, R (Eq. (9)), can be used to determine the dust particles is not controlled (ca. 106 particles per cubic feet
level of kinetic control over nanocluster formation [8]. in a typical laboratory [72]). A control experiment was done,
however, showing that increasing the amount of glass surface
growth rate k2 [A][B] k2 [B]
R= = = . (9) area (by adding glass beads) had no measurable effect on k1 or
nucleation rate k1 [A] k1 k2 [73].
Faster growth compared to nucleation, giving a higher value Since the initial reports of the 2-step, autocatalytic mecha-
of R, ensures that growth of existing nuclei dominates over the nism for nanocluster formation, Scheme 3, vide supra, a more
formation of new nuclei, thereby achieving near-monodisperse general mechanism for transition-metal nanocluster formation
(defined as ± 15% size distribution [7]) nanoclusters. The has been described which includes nanocluster agglomeration,
k2 [B]/k1 ratio is also a predictor of nanocluster size, smaller adding a third, B + B → C bimolecular agglomeration step [74]
values correlating with smaller nanoclusters and larger values and then, very recently, adding a fourth, autocatalytic agglom-
with larger nanoclusters [8]. eration step, B + C → 1.5C [66,75]. In these new steps, C
Interesting to note here is that the mechanism in Scheme 3 represents larger, approaching-like metal particles, Scheme 6.
can be used to give an excellent fit to the data obtained by While each of these modifications adds growth and agglomera-
Turkevich, Fig. 5 [70]. This is excellent, kinetic-based evidence tion steps to the overall mechanism, the nucleation step A → B
that the nanoclusters studied by Turkevich are formed via an remains the same. The 4-step mechanism has a distinctive ki-
netic curve (Fig. 6) in comparison to the 2-step mechanism in
2 Population balance kinetic modeling such as those done by Zukoski [14] Figs. 4 and 5. An example of the almost step-function like ki-
is another way to describe the more explicit treatment of all the steps given in netic curve is shown in Fig. 6; note the exceedingly good fit to
Scheme 5 herein, and Scheme 2 and Fig. 7 in a 2001 paper [69]. this unusually shaped kinetic curve.
which nucleation events are occurring—the formation of the

critical nucleus, according to classical nucleation theory [9,11],
as well as according to recent experimental results [78]. The
rate constant k1 is the nucleation rate constant, and the critical
nucleus is the size of the cluster at the end of the induction pe-
riod, when the clusters begin to grow and hence survive long
enough to become catalytically active. Measurement of the size
of the clusters at this point in the reaction should provide some
idea about the number of atoms in the critical nucleus, which
can then be compared to the number of atoms in the criti-
cal nucleus predicted by theory. Based on the GLC data for
the hydrogenation of cyclooctadiene, for the Ir0∼300 nanocluster
system an upper limit of the critical nucleus was estimated as
Ir0∼15 [8]. Recent results suggest that for the nucleation reaction
Fig. 6. Kinetic curve for the reduction of Pt(1,5-COD)Cl2 under H2 in ace- is actually nA → Bn and not 1A → 1B [67].
tone and concomitant cyclohexene hydrogenation. The curve is distinctive Interestingly, small values of the critical nucleus have been
in that no cyclohexene loss is observed for 0.5–3 h, and then suddenly be- found in other areas of science, for example in the protein
gins in an approaching step-function like manner. The excellent fit is to the agglomeration literature (see Appendix B), in which plotting
4-step mechanism in Scheme 5: k1 ≈ 1 × 10−7 h−1 , k2 ≈ 1.6 × 103 M−1 h−1 ,
k3 ≈ 1.7 × 102 M−1 h−1 , k4 ≈ 9.1 × 102 M−1 h−1 ; residual = 0.012. log(rate) vs log(concentration) suggests a critical nucleus size
of a single misfolded monomer [79]. This is a telling—if not
In the four-step, double autocatalytic mechanism, the rate frightening—insight, that a single misfolded protein is appar-
constants are determined using numerical integration, which ently involved in neurological disease [79].
involves finding the (ideally) global minimum in a five-dimen- The nanocluster composition studies establish the final oxi-
sional space (i.e., four rate constants plus the residual) [66]. dation state of the nanoclusters as zero, specifically Ir0∼300 [80].
As a result, the nucleation rate constant k1 carries with it an While these studies have provided a phenomenological nucle-
experimental uncertainty of 10±4 in the one well-studied case ation step of “A → B” (really “nA → Bn ”) with a nucleation
to date [66,75], despite there being ∼500 to 1000 data points rate constant k1,obs , the finer details by which the resultant
for the calculation of four parameters (rate constants), which metal0 nanoclusters are nucleated is not yet clear and is un-
gives ∼125 to 250 data points per parameter of high precision der further study. One unanswered question is: what steps does
data (±0.01 psig H2 ; ±0.025% out of 40 psig H2 initially) in the metal undergo in going from an ion in a salt to a zero-valent
that case [66,75]. (Recall that for the 2-step mechanism, vari- atom in a nanocluster? A number of hypotheses can be put for-
ability in k1 for the synthesis of Ir0n nanoclusters was limited ward, including:
to ± 101.2 [71] that seems to be mostly experimental in ori-
gin.) This 10±4 variability in the nucleation rate constant k1 • Under H2 reduction conditions, metal hydrides are a key to
for the 4-step mechanism has been addressed, and is at least in the nucleation process, Mx Hn+/−
y ;
part expected (as has been argued elsewhere [66,75]), given that • The reduction of metal ions and removal of ligands is
large-molecule materials are involved, in which less precision followed by nucleation of neutral, but highly energetic,
is a physical fact of Nature vs small-molecule chemistry [66]. M0 atoms, presumably highly ligated ones so as to be
In addition, sources of experimental error in the k1 and k2 meta-stable (see footnote 43(a) [62] and footnote 4(c) else-
and overall synthesis of Ir0∼300 nanoclusters have been exam- where [69]);
ined and include trace water, an unidentified impurity in the • The nucleation of metal ions, with some of their associ-
acetone solvent, and any source of heterogeneous nucleation
ated ligands, is followed by reduction of the overall cationic
[26,71,73]. The effects of always imperfect stirring have also
cluster, Mx Ln+
y ;
been shown to be a factor in the irreproducibility of any reac-
• Some combination of the previous hypotheses; the nu-
tion that involves autocatalysis [76]. All of these contribute to
cleus consists of a core of M0 atoms surrounded by metal
what Epstein has described as “kinetic indeterminacy” [77], and
shows why Oxtoby’s observation mentioned previously holds ions, [M0a Mn+ 0 n+
b Lc ] , possibly with hydrides, [Ma Mb Lc -
y+
true (i.e., his remark that “nucleation theory is one of the few Hd ]y+/0/− .
areas of science in which agreement of predicted and mea-
sured rates to within several orders of magnitude is considered Experimentally supporting or disproving these hypotheses is a
a major success” [9]). In short, the relatively large indetermi- major challenge in understanding the nucleation of transition-
nacy in k1 , for the 4-step mechanism of nanocluster nucleation, metal nanoclusters. There is currently disagreement in the lit-
growth, plus two kinds of agglomeration, appears to be a fact erature as to whether metal reduction to M0 , or critical nucleus
of Nature. formation to different species, occurs first [42,69]. While the
In the Finke–Watzky mechanism of slow nucleation and nucleation mechanism likely depends on reaction conditions,
autocatalytic growth, the induction period is the time during a more complete, more general picture of what happens during
the nanocluster nucleation and growth processes does not exist The Pt0 , Pt–H, and Pt20 intermediates were dismissed on the
at this time.3 grounds of high G of reaction (70.3, 42.0, and 64.1 kcal/mol,
The process of how nucleation occurs was considered in respectively). These arguments appear to be flawed, however,
some detail in the work of Finke et al. [62,69]. The high en- as they assume completely bare Pt atoms or ions without liga-
ergy of free Ir0 atoms is apparent when one considers that the tion from any of the other species present. It had already been
Hvaporization of Ir is 159 kcal/mol, which shows that free Ir0 pointed out in 1997 that bare M0n atoms are prohibited on ener-
is very unstable thermodynamically in comparison to bulk Ir0n getic grounds [62]. In other words, water molecules or chloride
formation; see footnote 44 elsewhere [62]. As first noted else- ions can (and likely do) bond to Pt ions and atoms, for example,
where [62,69], this suggests one of two possibilities: first, that thereby stabilizing them and lowering the G of the reaction.
if single Ir0 atoms are formed during the nucleation process Indeed, well-known molecular, M0 compounds with ligands
to make Ir0 –Ir0 bonds, then they must be stabilized by co- are stable and isolable, including Ru0 (1,5-COD)(1,3,5-COT)
ordination by solvent, anionic ligands (P2 W15 Nb3 O9− 62 in that (COT = cyclooctatriene) [82], Pt0 (1,5-COD)2 [83], Pt02 (dba)3
system [62]), or other ligands present [62], such as olefins. Al- [84], and Pd02 (dba)3 [84] (dba = dibenzylideneacetone).
ternatively, Ir–Ir bond formation may occur with at least one Ir The authors were left with the PtI mechanism, which they
atom still in a higher oxidation state (i.e., IrI –IrI or IrI –Ir0 ), fol- proposed as the true mechanism of Pt0n nanocluster forma-
lowed by reduction to all Ir0n . Restated, given that the formation tion. They admitted that this mechanism is “plausible, although
of free (i.e., unsolvated or unligated) M0 atoms in the nucle- not proven by direct experiment” [42]. (Of course no mecha-
ation process is thermodynamically disfavored, the remaining nism is ever proven; instead, alternative mechanisms are dis-
issues are: (i) when reduction takes place (as the final oxidation proven [85].) What we believe the authors meant is that they
state of the metal in the nanocluster is M0 [80]), and (ii) what were being careful to point out the lack of direct spectroscopic
are the roles of hydrides and other ligands in the nucleation evidence for PtI . Therefore, a PtI intermediate is a feasible, but
process. Overall, it seems likely that nucleation has more of not unequivocally supported, hypothesis in this case. Also mer-
an “inner-sphere” nature than is represented, for example, in iting further reinvestigation is a Ptn+/0 hypothesis.
2
purely “outer-sphere,” hard-sphere Monte Carlo simulation in a Prior to the above experiments, Henglein et al. reached a
2004 Nature paper [81]. similar conclusion for the formation of Pd0 colloids by radi-
olysis of PdII (NH3 )4 Cl2 [86]. (A 2006 review is available on
4.3. Henglein’s Pt, Pd, and Ag nanocluster formation studies
“Nucleation, Growth, and Properties of Nanoclusters Studied
by Radiation Chemistry” [87].) In this system, they used radi-
Henglein and coworkers have used thermodynamic argu-
olysis to form a hydrated electron that reduced PdII to PdI . They
ments and absorption spectroscopy to propose the formation of
then imagined two different scenarios for the formation of Pd
PtI intermediates in the aqueous reduction of [PtII Cl4 ]2− by hy-
“clusters” (e.g., Pd–Pd dimers, where the Pd atoms are coordi-
drogen in a 2000 paper [42]. They reported that aging the solu-
nated by NH3 , Cl− , and/or H2 O ligands). In one scenario, they
tions of [PtII Cl4 ]2− speeds the reduction when placed under H2 .
imagined two ligated PdI ions coming together to form Pd2+
This is presumably due to the displacement of two Cl− ligands 2
complexes. In the other scenario, two PdI complexes dispropor-
by H2 O, forming the more reactive complex PtII Cl2 (H2 O)2 ; the
diaqua complex has been found computationally to be a vi- tionated to Pd0 and PdII , and the complexed Pd0 atoms came
able intermediate (vide infra). Four separate mechanisms were together to form Pd0n . Although, as they stated, “a clear dis-
considered for the reduction of PtII and formation of Pt nano- tinction between these two cases cannot be made” [86], they
clusters based on the possible intermediates in the reduction: favored the Pd2+ 2 mechanism based on differences in reactivity
(i) formation of Pt0 by complete reduction by H2 , Eq. (10a), between the Pd “cluster” and fully reduced Pd colloids. Specif-
(ii) formation of PtI by partial reduction, Eq. (10b), (iii) Pt–H ically, the Pd2+ 0
2 cluster and Pdn colloids were each reacted with
−
H2 S and CN . In the reactions with H2 S, the Pd2+
hydride formation, Eq. (10c), and (iv) formation of Pt02 by the 2 reaction is
+
(Pd2 ) + 2H2 S → 2PdS + H2 + H , while the Pd0n colloid re-
2+
reaction of two PtCl2 (H2 O)2 molecules with H2 (Eq. (10d)).
action is Pd0n + nH2 S → (PdS)n + nH2 . In the reactions with
PtCl2 (H2 O)2 + H2 → Pt0 + 2H+ Cl− + 2H2 O, (10a)
CN− , the cluster formed Pd(CN)2− 4 , while the colloid formed a
2PtCl2 (H2 O)2 + H2 → 2PtI + 2H+ + 4Cl− + 4H2 O, (10b) different product, which was assigned as (PdCN)2 . Again, these
PtCl2 (H2 O)2 + 1.5H2 → Pt–H + 2H+ Cl− + 2H2 O, (10c) results are not solid evidence for or against either intermediate,
but they do demonstrate the feasibility of the Pd2+ 2 species as
2PtCl2 (H2 O)2 + H2 → 2Pt02 + 4H+ Cl− + 4H2 O. (10d) a participant in the nanocluster formation process. Problematic
in this work is that kinetic evidence for the proposed nucleation
3 As has been described previously [62], as well as back in Scheme 5, the steps is lacking. Nevertheless, the evidence for Mn+/02 —and/or
A + B → 2B step in the Finke–Watzky mechanism is in reality not Ir0 + IrI → M2 H0x we hypothesize—as a key species in nanocluster nucle-
Ir02 , but the more general Ir0n + IrI → Ir0n+1 , with n increasing as the nanocluster ation is slowly accumulating.
grows. The change in n is accounted for kinetically by a growth factor, xgrowth , Although Ag is not strictly a transition metal (having no
in the rate equation associated with the growth rate constant k2 . Therefore, the
key assumption made in this model is that the rate constant k2 for each of the
partially filled d or f orbitals), Henglein’s studies of Ag clus-
growth steps is the same, although it almost surely changes with the size of the ter formation are relevant to the present review because of the
nanocluster. In other words, the value obtained for k2,obs is an average value. mechanistic insights gained from those studies. The pulse ra-
Scheme 7. Reaction of Pt(acac)2 and Al(CH3 )3 to form the stable intermediate, Scheme 8. Balanced chemical reaction showing the formation of Pt(1,5-COD)-
which is presumed to go on to form Pt0 nanoclusters. (CH3 )2 from the stable Pt–Al complex and 1,5-COD.
diolysis method [87] was used to reduce Ag+ to Ag0 [88]. ciation of individual Pt0 atoms; that is, reduction of PtII to Pt0
UV–visible spectroscopy was used to detect and follow the for- occurs before nucleation. The authors did not give any evidence
mation of small silver dimers and tetramers, Ag+ 2 and Ag4 .
2+ for this; instead, they stated that “the rate controlling step for the
A mechanism for colloid formation was presented in which nucleation is the decomposition of the thermally unstable bin-
small Agx clusters form via coalescence of even smaller clus- uclear precursor molecules and is not the subsequent diffusion-
ters. These small clusters were designated “Type-I” nuclei [89]; controlled agglomeration of the single zero-valent Pt atoms into
though Henglein did not use this term, they can be thought of particles” (italics have been added) [94]. Therefore, they as-
as sub-critical nuclei. It was then assumed that at a critical size, sumed, without providing experimental evidence, that the inter-
the cluster grows by addition of a single Ag+ ion to the clus- mediate is reduced to give single Pt0 atoms, which then quickly
ter surface. These critical nuclei were called “Type-II” nuclei. It nucleate and grow into Pt0n nanoclusters. This assumption con-
was predicted that the critical size of Agm clusters “is possibly tradicts the thermodynamic considerations already described,
reached in the range of m from 10 to 20” [89] although no direct that bare M0 atoms are prohibited energetically [42,62].
evidence for this was given. The mechanism described by Hen- An interesting finding, in the work of Bönnemann et al.,
glein closely resembles the 1997 Finke–Watzky mechanism in is that if excess 1,5-cyclooctadiene is added to the reaction,
Scheme 3, in which nucleation of metal atoms up to the critical Pt0n nanoclusters are not formed. Instead, the dimethylplatinum
size is followed by growth via surface reduction of more metal complex Pt(1,5-COD)(CH3 )2 is formed, supposedly reforming
ions. Unfortunately, the kinetic data necessary to unequivocally Al(CH3 )3 (a balanced stoichiometry was not provided by the
support this was not published [89]. original authors; hence, a possible reaction stoichiometry is
The effects of added citrate ligand were examined in the for- given herein as Scheme 8) [96]. The 1,5-COD ligand apparently
mation of Ag clusters as well [89]. It was found that as the has a preventative effect on nucleation, breaking apart the Al-
bridged intermediate to form an organometallic complex. The
citrate concentration is increased, the particle size decreases
details of this not well-defined transformation, including kinetic
exponentially. The ability of the citrate to stabilize the size of
and mechanistic studies, or even a proven stoichiometry, have
the critical nucleus, Agm , was proposed to change with the cit-
not been reported.
rate concentration so that the size of the critical nucleus was
proposed to depend on the citrate concentration and its cluster-
4.5. Co nanoclusters from the Co0 complex Co2 (CO)8
capping action.
The formation of cobalt oxide nanoclusters has been stud-
4.4. Bönnemann’s Pt nanocluster intermediate studies ied in detail by Tannenbaum and coworkers [97]. The precursor
complex used in their studies was Co2 (CO)8 . The use of a pre-
The work of Bönnemann et al. on the preparation of Pt cursor in which the metal is already zero-valent nicely removes
nanoclusters offers another idea of what might happen dur- the issue of whether reduction occurs before or after cluster
ing nucleation [90–93]. In the reaction of Pt(acac)2 (acac = formation. However, the final product is the oxidized Cox Oy ,
acetoacetanoate) with Al(CH3 )3 , an intermediate was proposed so that an equivalent issue, of whether Co0 or oxidized Com+ n
consisting of two tetramethylplatinum species bridged by two clusters nucleate, is introduced into that work. The decomposi-
methylaluminum species, Scheme 7 [94]. This putative inter- tion of Co2 (CO)8 was assumed to take place via the well-known
mediate complex, Scheme 7, was characterized by 1 H NMR, process of removal of CO ligands before clustering occurs [98].
13 C NMR, mass spectroscopy, EXAFS, XANES, and anom-
In their work, Tannenbaum et al. proposed three steps to nano-
alous small-angle X-ray scattering (ASAXS) spectroscopy, cluster formation and stabilization in the presence of polymers
along with modeling using DFT. such as poly(methyl methacrylate) (PMMA) [97]: (i) the forma-
Although the proposed nanocluster intermediate was exten- tion of “metallomers” (small clusters—presumably the critical
sively characterized, the formation mechanism of the Pt nano- nuclei) by heating a solution of Co2 (CO)8 to 90 ◦ C, (ii) the ag-
clusters themselves from this putative intermediate was never gregation of metallomers to larger particles, and (iii) the adsorp-
investigated. Restated, lacking are the crucial kinetic studies tion of PMMA to the metal particle surface. The idea that the
necessary to demonstrate that this isolable (stable) complex is polymer does not interact with the metal throughout the nano-
a true intermediate along the reaction pathway, as opposed to cluster formation process, but instead waits in the background
a dead-end species as “Halpern’s Rules” for catalysis predict until a nucleus has formed and then grown to a cluster—even
(since it is stable and isolable) [95]. In addition, the authors when it is present in the reaction solution from the beginning of
assumed that nanocluster nucleation takes place via the asso- the reaction—is highly questionable and unsupported by firm
evidence. In this system, the main method of characterizing the

nanoclusters is infrared spectroscopy (IR), which was used to
follow the CO stretches of the precursor to measure the kinetics
of its decomposition. The IR data shows that no change in the
PMMA spectral features occurs until the onset of nanocluster
formation. This is contradictory to intuition, which predicts that
the polymer would interact with the metal from the beginning
of nanocluster formation. It also contradicts recent experimen-
tal evidence—albeit in a different, Ir system [99]—which shows
that especially lower MWave ∼ 3500 poly(vinyl pyrrolidone) Scheme 9. Reprinted with permission from Ref. [102].
appears to have a dramatic effect on the nucleation of Ir(0)n
nanoclusters [99]. (A significant issue in this second system is to nucleus sizes, but cannot accurately measure them. More
the broad molecular weight distributions of the polymers, so measurements taken after the induction period, but before the
that one does not know which MW component of the polymer end of the reaction, should provide more information about the
gives the observed effect [99].) The putative lack of PMMA formation of the nanoclusters, but such results have not been
interaction with the growing nucleus further stands in contrast reported.
to an earlier report studying Co nanocluster formation in the The formation of bimetallic “CoPt3 ” nanoclusters was re-
presence of polystyrene in which the authors concluded that the cently studied in detail by Weller et al. using the same
polymer was interacting with the “forming cobalt clusters” (ital- Co2 (CO)8 precursor along with Pt(acac)2 [102]. Because of
ics added) [100]. Because the PMMA is present with the precur- the lack of compositional studies, the nanoclusters should be
sor complex, it is much more likely that the polymer interacts written formally as “[Coa Pt3a (CO)b (acac)b ]n+/− ” to take into
with the metal throughout the nanocluster formation reaction. account all species that may be involved in the (unknown) nano-
Consistent with these points, it has recently been recognized cluster composition. This study focused on the dependence of
that “it is unlikely that these ligands only begin to function af- nucleation and growth rates on temperature and precursor con-
ter nucleation is complete” [101]. centrations. Using CNT arguments, specifically Eq. (4), they
Infrared data were also used to address the question of the explained why higher reaction temperatures give smaller nano-
sequence of oxidation and clustering of Co atoms [100]. The clusters by relating the final nanocluster size to the rate of
characteristic CO absorption band at 1858 cm−1 and Co–O nucleation, Scheme 9, stating without evidence that the nucle-
band for the oxidized (Co2 O3 )n clusters at 883 cm−1 were ation rate is more sensitive to temperature than the growth rate.
both followed during the reaction. The appearance of the Co– These authors also observed that, with an increase in the con-
O band did not correlate with the decrease in intensity of the centration of their nanocluster stabilizer, 1-adamantanecarbox-
CO band. The decrease of the CO band, corresponding to the ylic acid (ADA), the final nanocluster size increased, a result
decomposition of Co2 (CO)8 , began immediately, while the ap- opposite to Turkevich’s finding with Au/citrate described earlier
pearance of the Co–O band, corresponding to the formation of [65]. The stabilizer is hypothesized to interact strongly with the
Co2 O3 clusters, showed a ∼2 h induction period. This led the monomers, slowing nucleation and leading ultimately to larger
authors to conclude that decomposition of Co2 (CO)8 occurs to nanoclusters, as shown in Scheme 9 [102].
form Con nuclei, which are then oxidized to yield (Co2 O3 )n The nucleation mechanism put forth by Weller et al. is as
clusters. A major problem with these otherwise valuable stud- follows, although the kinetic data necessary to put the follow-
ies is that CO loss from Co2 (CO)8 is a significant part of, if ing mechanistic speculation on firm ground are lacking [102].
not the, rate-determining step. This, unfortunately, hides kinet- The critical nucleus for formation of the Co/Pt nanocluster is
ically the nanocluster nucleation and growth steps of interest. proposed to consist of only Co atoms (i.e., and no Pt) based on
The Co2 (CO)8 thermolysis system, therefore, is probably not the observation that an increase in Pt led to an increase in the
the ideal one in which to learn the intimate details of the nucle- final size of the CoPt3 particles. One can suppose that the nu-
ation and growth of (Co2 O3 )n nanoclusters. clei are composed of Con (CO)m units, where n and m are small
The size of the critical nucleus was estimated for the integers; experimental evidence for or against this hypothesis
Co2 (CO)8 system using TEM and dynamic light scattering is lacking, however. The Co clusters are then hypothesized to
(DLS) experiments. In these experiments, samples from the catalyze the reduction of PtII from Pt(acac)2 to PtI , which is in-
reaction solution were analyzed during the induction period corporated into the Co/Pt cluster. The proposition of this PtII
for the formation of Con nuclei, as well as at the end of the to PtI reduction step is based on computational studies of Pt
(Co2 O3 )n nanocluster forming reaction. No clusters were ob- cluster formation (vide infra). The cluster is then thought by
served during the induction period above the DLS detection the authors to be reduced by either CO or 1,2-hexadecandiol,
limit of 3 nm; therefore, an upper limit to the critical nucleus which is present in the reaction solution. Finally, the authors hy-
size of 3 nm was assigned to the Con nuclei which in turn can pothesize that the process of PtII → PtI reduction and addition,
be used to calculate a Co1285 cluster in the limit of a hypo- followed by cluster reduction, is iterated to form fully reduced
thetical, unligated 3 nm Co particle. This study illustrates the CoPt3 alloy nanoclusters.
disadvantage of TEM (and DLS) mentioned earlier, that rela- The mechanism put forth above was formulated based on
tively high detection limits are only able to set an upper limit CNT and the sizes of the nanoclusters, the latter with respect
growth, although this is not certain. The necessary rigorous ki-

netic analysis of this UV–visible data is missing. Again, the
balanced reaction stoichiometry required for reliable mechanis-
tic studies is also missing.
4.7. Observation of Rh4–6 clusters in dehydrogenation

reactions by XAFS
In 2005, Chen, Fulton, Linehan and Autrey at Pacific North-

west National Labs (PNNL) reported an EXAFS and XANES
study on the rhodium-catalyzed dehydrogenation of dimethy-
lamine borane beginning with a [Rh(1,5-COD)Cl]2 catalyst
precursor [103]. This reaction, which shows promise for ap-
plications of hydrogen fuel storage and release [104], had been
Fig. 7. Absorbance profiles of Pd2+ ions (top) and Pd0 clusters (bottom). Three previously studied by Manners’ group [46,105] with the goal of
distinct periods can be seen. During the flat parts of the Pd2+ profile, reduction determining the true catalyst. Manners’ studies indicated that a
of Pd2+ stops, but cluster growth is believed to be still occurring at these times. heterogeneous Rh species, likely a Rh(0) colloid, catalyzes the
Reprinted with permission from Ref. [43]. dehydrogenation. Using XAFS, the PNNL team studied the de-
hydrogenation of dimethylamine borane starting with [Rh(1,5-
to the dependence of nanocluster size on the nucleation and COD)Cl]2 . Taking advantage of the strong Rh backscattering
growth rates. However, without the support of the necessary in the EXAFS, they were able to show that small Rhn+ x clus-
kinetic data, the conclusions in the study can be, strictly speak- ters are formed in the reaction—suggesting, but not kinetically
ing, only considered hypotheses based on nucleation theory verifying or disproving—that these small, sub-nanometer (ca.
(which has been shown to be inaccurate in many cases, vide 0.3 nm) clusters and not larger Rh(0) nanoclusters, are the true
supra). The lack of an established composition for the resultant catalysts. These sub-nanometer clusters were initially assigned
[Coa Pt3a (CO)b (acac)c ]n+/− , as well as a balanced reaction sto- to be Rh6 based on comparisons with the carbonyl complex
ichiometry that is the first step to reliable mechanistic work, are Rh6 (CO)6 ; since this report, further studies have suggested that
other, fundamental problems with this work. the clusters are actually Rh4–6 [106], results which emphasize
the 20–30% or more error in determination of the coordina-
4.6. Rothenberg’s UV–visible study of Pd, Au, and Ag tion number (i.e., and thus the number of metal–metal bonds)
nanocluster formation by EXAFS, a point emphasized by Gates [107].
Examination of previous literature reveals at least two re-
As mentioned earlier, a problem with in situ UV–visible ports that support the idea of a M4 , and we hypothesize more
measurements of nanoclusters is that many other species are likely a M4 H4 , sub-nanometer cluster, as an important cat-
present which could interfere with the spectrum of the nano- alytically active, species. In 1987 it was hypothesized by Duff
cluster. To deal with this problem, Rothenberg et al. devised et al. that nucleation occurs by addition of metal atoms to a
a system, called the “net analyte signal/principle component tetrahedral nucleus, since the tetrahedron is the simplest meta-
analysis” (NAS/PCA), in which the spectra of the interfering stable unit [108]. TEM visualization of silver clusters provided
species are essentially subtracted out of the spectrum, leaving evidence for this hypothesis in that a structure reflecting the
a difference spectrum that they assume is only the spectrum symmetry of an underlying Td species is seen; computer-image
of the nanoclusters [43]. These authors used the NAS/PCA simulations corroborated this evidence. However, as described
method to study the formation of Pd, Au, and Ag nanoclusters, above, this evidence must be considered equivocal as TEM is
as well as to study the possibility of forming bimetallic nano- capable of affecting the observed structure of especially non-
clusters using combinations of those metals. Following the cor- third-row transition-metals [46,47].
rected nanocluster spectrum and the metal salt spectrum with In 2001, the hypothesis of polytetrahedral organization of
time, they were able to measure the kinetics of salt reduction particles was put forth by Chaudret et al. for the formation of Rh
and nanocluster formation, Fig. 7. From their data, they con- nanoclusters from reduction of [Rh(C2 H2 )2 Cl]2 by Cp2 V [109].
cluded that reduction of the metal salt occurs before the for- In this case, both high-resolution TEM and wide angle X-ray
mation of clusters. Also, for the formation of Au0 nanoclusters scattering (WAXS) were used to support the tetrahedral cluster
from HAuIII Cl4 , a AuI intermediate was inferred based on the hypothesis, but lack of a defined stoichiometry or any kinetic
NAS/PCA data and the reduction potentials of AuIII and AuI . evidence makes the conclusions equivocal.
The authors noticed distinct induction periods in which no de- These data support the hypothesis that the critical nucleus
tectable clustering took place (see Fig. 7). Although they do is not composed of neutral metal atoms, but decorated with
not mention the implications of this, there is a possibility of us- ligands—again, we suspect that hydrides are likely more com-
ing this data—in particular the first induction period—to gain mon ligands. Further evidence of the possibility of Rhn+ Hn
information about the critical nucleus of the clusters. The sec- sub-nanometer clusters comes from the literature of Rh-hydride
ond and third clusterings are presumably periods of nanocluster cluster complexes, for example [Rh(1,5-COD)H]4 [110] and
(a)
Fig. 8. Analysis of Hirai’s [115] transition-metal nanocluster data showing a
correlation between Hvap and nanocluster size for six different metals.
[Cp*RhH]2+ 4 [111]. Computational evidence also supports the

idea of non-zero-valent metals as crucial intermediates en route
to nanoclusters (see Section 6). In light of the PNNL group’s
EXAFS results, the idea of M4 (or M4 H4 as we prefer at
present) sub-nanometer catalysts deserves immediate attention
and may in fact open up a new area of active, sub-nanometer
catalysts. Tantalizingly consistent with the idea of a M4 sub-
nanometer catalyst is the observation of Finke, Finney, and
Starkey-Ott [67] showing that the kinetic critical nucleus of
the well-studied Ir(0)n nanocluster system is n 2. Of course,
2 × 2 = 4, so that dimerization (aggregation) to a more stable
Ir4 nucleus is one intriguing idea requiring EXAFS, XANES,
(b)
and other, in-progress studies [112].
Fig. 9. Analysis of Bönnemann’s [117] transition-metal nanocluster data for the
5. Other factors affecting nanocluster preparation preparation of nanoclusters in THF (A) and in toluene (B). The slope of the line
in (B) is an order of magnitude greater than that in (A). This could be due to the
different metals studied, the different solvent, or some other factor(s).
Control over nanocluster nucleation is expected to lead to
control over nanocluster size, shape, and other properties. In
this section, control over nanocluster nucleation is examined in strength, as reflected by a higher value of Hvap , should favor
two ways: the first looks at the inherent effect of the transition nucleation over growth and result in smaller nanoclusters. Con-
metal itself on the size of the critical nucleus, and the second versely, a metal with lower M–M bond strength and therefore
examines the technique of seeding nucleation by adding pre- lower Hvap should favor growth over nucleation, resulting in
formed clusters. larger nanoclusters. All of this, of course, assumes that nano-
clusters of all metals are formed under the same mechanism—a
5.1. Effect of metal–metal bonds on the critical nucleus necessary hypothesis for the plots which follow, but not one ex-
pected to be completely correct. Restated, this hypothesis can
As described earlier and shown in Fig. 1, the critical nucleus only be supported or disproven using identical reaction con-
has a specific size, r ∗ , associated with it. The hypothesis can ditions for all metals (temperature, concentration, solvent, and
be put forth that for transition-metal nanoclusters, the size of ligands).
the critical nucleus is dependent on the strength of the M–M An examination of literature preparations of transition-metal
(metal–metal) bonds (see Figs. 8, 9, and 10 herein, vide in- nanoclusters using many different metals seems to support the
fra, for initial data supporting this hypothesis). Precedent for hypothesis: that the size of the nanoclusters in general de-
this hypothesis can be found in Stranski’s 1935 paper, where creases with increasing heat of vaporization, ΔHvap . A well-
he uses a Hvaporization(sublimation) vs a Hsurface energy term known relationship that must be mentioned here is the Gibbs–
as part of his theoretical treatment [113]. Since a metal’s heat Thompson relation, stating that the partial molar free energy of
of vaporization, Hvap , can be used as a measure of the M–M a metal atom in a particle of radius R, μ(R), differs from that
bond strength, and since the critical nucleus strongly influences in the bulk, μ(∞), as a function of the surface free energy of
the final size of the nanocluster, there should therefore be a that same metal, γ , and the volume per metal atom of the bulk
correlation between a metal’s Hvap and the size of the metal metal, Ω, Eq. (11). An excellent recent paper which discusses
nanocluster. More specifically, a metal with higher M–M bond the Gibbs–Thompson relation, and shows nanocluster size ef-
A simple experiment to determine if the solvent is the (major)

factor that determines the slope of the line would be to reduce
the same set of metal precursors in different solvents. The phys-
ical meaning of the slopes of the lines could provide key insight
into the nature of these two domains of nanocluster size depen-
dence on M–M bond strength.
The previously unpublished results in Figs. 8 and 9 are quite
interesting. However, what is needed to determine if such a cor-
relation between nanocluster size and a given metal is in fact
real and perhaps more general are studies in a system in which
the nanocluster formation mechanism is known and where all
the other conditions besides the metal are kept constant. The
nanocluster system in which the autocatalytic A → B, A + B →
2B mechanism has been extensively studied is in principle the
Fig. 10. “Two-point plot” for the reduction of [(1,5-COD)Ir·P2 W15 Nb3 O62 ]8+ best system at present in which to perform the needed studies
and [(1,5-COD)Rh·P2 W15 Nb3 O62 ]8+ . The Ir and Rh complexes are the only [62]. The separation of nucleation and growth k1 and k2 steps
ones that have been successfully synthesized and reduced to obtain nano- should allow each of those steps to be correlated with nano-
cluster sizes, but suggest that a correlation may be present between metal
cluster size and M–M bond strength.
Hvaporization and nanocluster size in this system as well.
The needed studies were attempted for the polyoxoanion-
stabilized transition-metal nanoclusters, [(1,5-COD)M·P2 W15 -
fects even more dramatic than those predicted by Eq. (11), is a Nb3 O62 ]n+ . Unfortunately, only the Ir [62] and Rh [118] pre-
2002 paper by Campbell and coworkers [114]. cursor complexes have been successfully synthesized; the nano-
2γ Ω cluster size vs Hvap “2-point plot” for these metals is shown
μ(R) − μ(∞) = . (11)
R in Fig. 10. At least its trend is as expected, but, of course, the
The most illustrative data consistent with the preceding hy- “2-point line” is unconvincing. This in turn led to the idea of us-
pothesis is Hirai’s classic preparation of nanoclusters from ing the commercially available metal salts [MI (1,5-COD)Cl]2
transition-metal chlorides (RuCl3 ·3H2 O, RhCl3 ·3H2 O, PdCl2 , (M = Ir, Rh) or [MII (1,5-COD)Cl2 ] (M = Pt, Pd, Ru) under H2
H2 PtCl6 ·6H2 O, and HAuCl4 ·4H2 O), one oxide (OsO4 ), and in propylene carbonate as the solvent (as propylene carbonate,
one nitrate (AgNO3 ) [115]. A plot of nanocluster size vs the with its relatively high dielectric constant of 69, can allow even
metal Hvap is shown in Fig. 8, and gives a roughly lin- weak nanocluster stabilizers such as Cl− to give stable, soluble
ear plot with an R 2 value of 0.915 [115]. The nanoclusters nanoclusters [118,119]). Optimization of the experimental con-
were prepared by refluxing solutions of the precursor com- ditions to allow this direct comparison of nanocluster size vs
plexes and polyvinyl alcohol (PVA) in a methanol/water mix- metal is currently in progress. Also fascinating here is the idea
ture in air. The sizes of the nanoclusters were measured by of trying to see the critical nucleus, and other nanocluster size
TEM; unfortunately, this was the only characterization method vs time data, by EXAFS and XANES plus the requirred kinetic
used so the true composition of the nanoclusters is unknown studies, efforts that are in progress [120].
(and, for example, since the nanoclusters were formed in air,
the formation of surface oxides is highly likely [116]). Also, 5.2. Seeded nucleation
the methanol used as a solvent can form formaldehyde and
formic acid, both of which may interact with the metal sur- Growth of nanoclusters using seeds as nuclei has been per-
face. This gives a possible total composition of the nanoclusters formed since 1906 [121]. In an important study by Buhro et al.,
of [M0n (Cl− )m (PVP)p (O)q (OH− )r (OH2 )s (CH3 O− )t (CH2 O)u seeds of Au101 (PPh3 )21 Cl5 were used to nucleate nanoparticles
(HCO− 2 )v ]
x+/− , with unknown of values of m–v. of Bi, In, and Sn [122]. It was found that using a small con-
Another example showing a correlation between nanocluster centration of Bi, In, or Sn gave quite narrow size distributions
size and Hvap comes from our analysis, Figs. 9a and 9b, of (from 5.6% to 13.2%). The experimental average nanoparticle
Bönnemann’s 1996 report of “nanoscale colloidal metals sta- size matched calculated sizes with ratios of observed to theoret-
bilized by solvents and surfactants” [117]. In this preparation, ical sizes at or near 1.0 [122]. In the absence of the Au seeds,
halide salts of Ir, Pd, Fe, Rh, Ru, and Co were reduced in THF, “extremely broad size distributions” were observed (no quanti-
and salts of Cu, Ag, Au, and Cr were reduced in toluene. Since tative data were provided, however [122]). If the concentration
the metals were prepared in different solvents, two different of Bi, In, or Sn was too large, self-nucleation of the metal be-
plots are shown in Fig. 9, one for each solvent used. The corre- came competitive with the seeded nucleation and broader size
lations are at least suggestive—for THF, the R 2 value is 0.856 distributions resulted which, in turn, did not match the theoret-
and for toluene the R 2 value is 0.759 [117]. The difference in ical sizes. The advantage of the seeding method is the obvious
slopes between the two curves is very interesting (in curve A control over nucleation and the resultant size and size disper-
the slope is −0.021 nm/kcal mol−1 , and in curve B the slope is sion. Disadvantages of this method include that the resulting
−0.334 nm/kcal mol−1 ) and could be due to the different sol- Bi, In, and Sn nanoclusters are contaminated with Au, although
vents used, the different metals, both of these, or other factors. the Au is at least initially at the center of the nanocluster and
its amount is small (a few percent), depending on the size of effects of pressure and temperature on nucleation rate and par-
the nanoclusters. Still, if 100% pure metal nanoclusters are re- ticle concentration. In contrast, relatively few computational
quired, this heteroelement-seeding procedure would need to be studies exist for the formation of transition-metal nanoparticles
avoided. in solution.
Self-nucleation has been reported as a problem to be avoided Recently, Ciacchi et al. published computational studies us-
in the work of Murphy et al. [123]. They found that reducing ing first principles molecular dynamics simulations of the nu-
AuCl− 4 in the presence of preformed 12 ± 2 nm Au nanoclusters cleation of Pt clusters from [PtII Cl4 ]2− in aqueous solution
gave varying sizes and size distributions of Au nanoclusters [134], and their growth by addition of PtII complexes [135]. In
depending on the ratio of seed to Au salt and on the rate of their simulations, they observed the formation of PtI –PtII and
addition of reducing agent (ascorbic acid in that study). Specif- PtI –PtI dimers, which could then react with another PtII species
ically, when the concentration of AuCl− 4 is much larger than to form PtI –PtII –PtI or PtI –PtII –PtII trimers [136]. Based on
the concentration of Au nanocluster seeds, self-nucleation of their observations, the authors proposed that at least in the case
the AuCl− 4 dominates over seeded growth (just as Buhro et al. of platinum, cluster formation occurs via bonding of metal ions
found for the preparation of Bi, In, and Sn nanoclusters [122]). before reduction to Pt0 , as opposed to full reduction to Pt0
Self-nucleation was also observed by Murphy et al. when the re- followed by nucleation of the Pt0 atoms. Their proposed nu-
ducing agent was added quickly [123]. Slow addition of ascor- cleation mechanism does away with the classical idea of the
bic acid prevented self-nucleation, but resulted in deviations in critical nucleus; instead they state that “each PtII complex that
nanocluster shapes from spheres. The authors presented size- reacts with a reducing electron can be thought of as a ‘critical
separation and X-ray powder diffraction data to disprove the nucleus’ for the growth of metallic platinum clusters” [136].
hypothesis that very small nanoclusters (i.e., too small to be The discarding of the well-entrenched idea of the critical nu-
observed in the TEM) were formed and acted as (hidden) seeds cleus, based on this one computational study—and its lack of
for the nanocluster formation [123]. analysis or discussion of the extensive literature cited herein
Seeded nucleation has been found to be necessary for the re- on the critical nucleus concept—is at best unjustified. In fact,
duction of some complexes. Whitesides found that the platinum these computational observations are in line with experimental
complex Pt(1,5-COD)(CH3 )2 cannot be reduced by H2 unless considerations concluding that partially oxidized metal clusters
Pt(0) is present as a catalyst [124]. More recently, Finney and are formed before the critical nucleus is attained. It would be
Finke found that the presence of Ir0∼300 nanoclusters as seeds of interest to return to these studies and see if Pt20 , Ptn+/0
4 or
catalyzed the reduction of Pt(1,5-COD)(CH3 )2 under H2 [125]. Pt4 Hn+/0
4 satisfy the criteria of being either critical nuclei or
In this case, the formation of putatively Ir0 (core)/Pt0 (shell) special (metastable) intermediates.
nanoclusters quickly agglomerated to bulk metal. As with In their calculations, the authors assumed a PtI intermedi-
Buhro’s system, this technique results in bimetallic nano- ate, and accordingly ran their simulations with the [PtII Cl4 ]2−
clusters and is, therefore, not desirable if very pure nanoclusters ion and one reducing electron; that is, they simulated their
are needed—or very desirable if Ir0 (core)/Pt0 (shell) bimetallic specific mechanistic hypothesis and found it acceptable ener-
nano-onions [8] are one’s goal. getically. Therefore, while their calculations are evidence for
The use of nanoclusters as seeds for organic structures the plausibility of a mechanism involving PtI species, they do
has also been reported [126]. Specifically, rods of arachidic not provide the necessary disproof of other mechanisms [85]
acid (CH3 (CH2 )18 COOH) have been nucleated onto CdSe (e.g., including complete reduction to Pt0 as a significant step
nanocrystals by spin coating. The rods are of uniform height on the process, or the Pt2+/0
n , Pt4+/0
n or Pt4 H4+/0
n alternative hy-
(0.95 ± 0.09 nm) and width (5.39 ± 0.05 nm) and lengths that potheses put forward above). These computational studies are,
vary from 50 to 250 nm. These results suggest that the use in general, based on assumptions about what happens during
of nanoclusters as seeds will be useful not only for the for- nucleation; more solid experimental evidence of what occurs in
mation of other nanoclusters, but also for the preparation of the nucleation events is needed to guide further computational
organic–inorganic hybrid materials [127]. The ability to tune studies, to add hydrides or other appropriate ligands to them,
the properties of the nanocluster seeds will in turn allow for the and overall to make such calculations more realistic and, there-
ability to change the properties of the hybrid materials. fore, even more valuable.
6. Computational studies of nanocluster nucleation 7. Conclusions
The vast majority of computational work on the nucleation The main conclusions of this review are:
of clusters has been done for gas-phase nucleation of liquid
droplets or solid nanoparticles formed by chemical vapor de- • There is currently no clear, experimentally quantitative
position, since computations for gas phase systems are, of description of what happens during the nucleation of
course, much simpler than for, say, solution nanocluster sys- transition-metal nanoclusters in solution. Indeed, the nucle-
tems. This is illustrated by the extensive computational studies ation of clusters in general is poorly quantitated in general,
by Zachariah et al. of the nucleation, growth, and agglomeration due mainly to a lack of good quantitative experimental ki-
of Si [128–130] and SiO2 nanoparticles [131–133] by CVD. netic and other data. The kinetics of reactions of known
Simulations of the nucleation of SiO2 particles focused on the stoichiometry are especially lacking in all but a few cases.
• Although nucleation and growth of clusters has been stud- dots of the form CdE (E = Se, S, Te), with the most attention
ied exhaustively in the theoretical sense, experimental re- given to CdSe. Although these are not transition-metal nano-
sults are far less common and in many cases contradict clusters, they have been the subject of significant study with
theory, often by many orders of magnitude. regard to the mechanism of their formation and, therefore, war-
• A number of issues remain to be investigated with regard rant at least brief mention here. The CdE nanocrystals are pre-
to the nucleation of nanoclusters, including: pared using the hot-injection method described by Bawendi et
◦ If the formation of the nuclei occurs before, during, or af- al. [138], in which a chalcogen compound, often in a solution
ter the reduction (or oxidation) of the metal in the metal of phosphine, is injected into a hot (300 ◦ C) solution of a
precursor complex. Cd2+ species (usually Cd(CH3 )2 or CdO) in a phosphine so-
◦ If ligands—and especially hydrides to start—are a key lution. After the injection, the solution is allowed to cool to
part of nucleation, or are they merely spectators until the <100 ◦ C. This change in temperature from the hot injection to
nanoclusters are fully formed, as has been suggested, the cooling immediately afterward accomplishes the separation
albeit without evidence [97]. Or, are ligands involved of nucleation and growth that results in narrow size distributions
in every step of nanocluster formation, including nucle- of the nanocrystals, impressively as low as 5–6% as measured
ation, as Finke and coworkers [8], as well as van Embden by TEM [138]. Nucleation occurs over a ∼20 s timescale and
and Mulvaney [101] have suggested, as the influence of growth is slower, occurring over the next ∼60 min. It is in-
PVP over nucleation, for example, show [99], and as teresting to note that the relative rates of (fast) nucleation and
one’s intuition expects? (slow) growth for this system is opposite that of the Finke–
◦ If ligands are crucial to the nucleation step how does Watzky mechanism [62], in which slow nucleation is followed
the strength of the metal–metal bond (or metal–hydride by fast, autocatalytic nanocluster growth. It should be remem-
bond) affect the critical nucleus and, therefore, the final bered that the CdE nanocrystal preparations are performed at
size of the nanoclusters, and how do the size and shape high temperatures, as opposed to the room-temperature synthe-
of the critical nucleus relate to those of the fully formed ses of transition-metal nanoclusters; this could be one source
nanoclusters? Much of the experimental evidence sup- of the apparently different mechanisms (the systems are com-
ports the hypothesis that nuclei consist of at least par- pletely different as well, of course).
tially oxidized metal clusters which are reduced after The kinetics and mechanism of CdSe growth were first stud-
reaching some critical size, although the details of this ied by Alivisatos et al. in 1998 [139]. These authors focused
mechanism have yet to be worked out. their study on nanocrystal growth and avoided studying nu-
• The Finke–Watzky 2-step mechanism fits nanocluster for- cleation, noting that “the kinetics of nucleation are difficult
mation, as well as solution agglomeration of proteins in at to study” [139]. Photoluminescence spectroscopy was used to
least some cases (see Appendix C). As such, evidence is probe the growth of the nanocrystals.
accumulating that the 2-step mechanism may be the more Qu et al. presented a possible in situ method for character-
general, simplest (Ockham’s Razor) mechanism of at least izing CdSe nanocrystals by UV–visible spectroscopy to probe
some 1st order phase transitions in solution, a hypothesis the nucleation and growth of the nanocrystals [5]. In this way,
under further investigation [137]. the samples can be characterized at the actual reaction tempera-
• Promising are the studies indicating a higher order kinetic ture instead of taking aliquots of reaction solution, cooling them
critical nucleation step, nA → Bn [67] and the EXAFS to room temperature, and then characterizing them [139]. This
and XANES studies implying a Rh4 (or Rh4 H4 ) meta- avoids problems of changing the reaction solution by removing
stable (sub-) nanocluster [58,112]. Further efforts on the significant amounts for characterization, as well as changing
M4 /M4 H4 critical nucleus hypothesis, detailed herein for the nanocrystals themselves by cooling them before characteri-
the first time, are being actively pursued [58,112]. zation. The in situ UV–visible spectra of the solutions at 250 ◦ C
were indeed different from those taken at 25 ◦ C; specifically,
The ultimate goal of nucleation studies, as well as studies of absorbances were shifted to higher wavelength and broadened
growth and agglomeration, of transition-metal nanoclusters is in the spectra taken at higher temperatures. Unfortunately, the in
a general, complete mechanistic understanding of what occurs situ UV–visible method was abandoned in favor of photolumi-
during nanocluster formation. The information gained from nescence spectroscopy [139] along with computer simulations
such an understanding should shed considerable light on the to measure the sizes of the nanocrystals. Therefore, while the
process of nanocluster formation, therefore leading to enhanced method of in situ UV–visible spectroscopy was shown to be
control over nanocluster size, shape and other desired proper- easily employed, the only conclusion based on the UV–visible
ties. It is hoped that the present review will assist in achieving method given in this study was that the spectra are different at
these goals. higher temperatures. No information about the nucleation of the
nanocrystals was obtained using UV–visible spectroscopy.
Appendix A. Nucleation studies of semiconductor It is notable that Qu et al. found that classical nucleation the-
nanocrystals ory does not accurately describe CdSe nanocrystal nucleation,
at least the “semiquantitative” data collected by the authors [5].
Of the nanoparticle systems formed in solution, one of the Specifically, these authors found that the concentration of the
most thoroughly studied systems is the formation of quantum CdSe nanocrystals depends on the concentration of monomers
in solution, while CNT predicts that there should be no de- nanocrystals are determined by the absorbance, and the concen-
pendence. (Note that in the recently observed nucleation step tration of particles is determined by the absorbance. Accurate
nA → Bn for transition-metal salts, the concentration of nano- determinations of the sizes and concentrations of the particles
clusters is also dependent on the concentration of “monomer,” depend on good calibration curves; this has been done by at
[A]—this dependence is manifested as k1,obs = nk1,true [A]n−1 least two researchers [140,141] and results have been found to
[67].) The various uncertainties in the CdSe nanocrystal forma- agree with both determinations [142].
tion data, however, led the authors to treat with caution their
conclusion of a nanocluster dependence on monomer concen- Appendix B. Solid-state nucleation studies
tration. In particular, as noted by the authors [5], the uncertainty
and inherent irreproducibility in the accuracy of the injection The solid-state chemical literature has a history of nucleation
rate of Se into the Cd solution reduces the accuracy of, and thus and growth studies dating back to the 1930s, and the kinetics of
the confidence in, the measured nucleation data. solid-state reactions have been reviewed [143,144]. The reason
The nucleation of CdSe nanocrystals was recently studied for including these studies here, as well as those on protein ag-
in detail by van Embden and Mulvaney [101]. Like Qu et glomeration in Appendix C, is that it is probable that nucleation
al. [5], they found that CNT is unable to explain the nucle- and growth phenomena throughout nature are closely intercon-
ation of CdSe; the slow nucleation they observed also did not nected, so that insights from one area can fuel advances in other
agree with LaMer’s “burst” colloid nucleation mechanism [59, areas.
60]. These authors formulated the hypothesis of “ligand con- A brief summary of the solid-state decomposition mech-
trolled nucleation,” in which certain ligands, referred by the anism literature is given in Table B.1. Solid-state nucleation
authors as “nucleation agents,” will have an affect on the size theory has its roots in CNT,
of the critical nucleus, as well as the concentration of nuclei
formed and their probability of surviving. To test this hypothe- G = mGB + σ γ , (B.1)
sis, they added the ligand bis(2,2,4-trimethylpentyl) phosphinic in which the free energy change depends on the shape of the nu-
acid (TMPPA) to a reaction solution containing a modifica- cleus, as expressed by the shape factor σ (4πr 2 for a spherical
tion of the general preparation of CdSe nanocrystals [138]. The interface) and the strain energy per unit area of the interface, γ
TMPPA ligand was chosen with the idea that it would preferen- [143]. In this equation, m is a function of the nucleation area
tially bind to the CdSe nanocrystal surface over the monomers, (4/3πr 3 for a spherical nucleation region) and GB is the free
preventing redissolution of the nuclei. Using UV–visible spec- energy difference between the phases.
troscopy, the authors found that when TMPPA was added to the The basic starting equation for solid-state nucleation and
reaction, smaller CdSe nuclei resulted, increasing the concen- growth is the Avrami–Erofe’ev equation, [− ln(1 − α)] = (kt)n ,
tration of nuclei vs the control reaction without added TMPPA, where, according to Erofe’ev’s derivation [148], α is the frac-
and slowing the nanocrystal growth process. It was inferred that tion of starting material that has reacted to form product, n is a
the binding of the ligand to the nucleus lowered the free energy parameter that can range from 2 to 4 in so-called “accelerato-
of the critical nucleus and the surface free energy, γ ; the result ry” kinetics or can be 1, 1/2, 1/3, and so on in “deceleratory”
is a lowering of the barrier to nucleation. The relevant equations kinetics, and k is the rate constant. A problem appears imme-
used (Eqs. (A.1) and (A.2)) are similar to those given above by diately for chemical systems with regard to the meaning of the
Auer and Frenkel [19] (Eq. (6)): parameters n and k. First, the notion of “acceleratory” vs “de-
celeratory” kinetics is not well defined, and in some cases the
Gmax
knucleation = aT exp − , (A.1) parameter n is allowed to change during the course of a single
RT
experiment. Therefore, this parameter seems to be little more
16πγ 3 than a fitting factor, included to ensure a good fit without adding
Gmax = . (A.2)
3Gv any physical meaning. Second, the rate constant k appears to be
Equation (A.1) has the general form of an “Eyring-type” a conglomerate of nucleation and growth rate constants, convo-
equation. In Eq. (A.2), Gv is the free energy per unit volume luted into a single constant [157].
for the condensation of the monomer, Gv = −(RT /Vm ) ln S, As can be seen in Table B.1, the Avrami–Erofe’ev equation
where Vm is the molar volume of CdSe (in this case) and S is has been generalized several times since its initial derivation,
the degree of supersaturation; that is, the product [Cd][Se] in that is, all of the subsequent models simplify to the Avrami–
solution divided by the solubility product [Cd]eq [Se]eq . The ul- Erofe’ev equation under certain conditions. Basically, fitting
timate utility of the addition of “nucleating agents,” according solid-state kinetic data involves choosing a general expression
to the authors, is in the ability to tune the nucleation and growth from one of those in Table B.1 (or deriving a more complex
of the nanocrystals (in conjunction with the addition of “growth one), and then changing parameters until a sufficiently good fit
agents”), leading to a control of nanocrystal sizes. is attained. In the majority of reports, the values of the relevant
The kinetics of nucleation (and growth) of the CdSe nano- parameters are given, but their chemical meaning is generally
crystals are measured with UV–visible spectroscopy. The not discussed. In fact, worth noting is that most of the mod-
nanocrystals absorb between ∼500 and 700 nm depending on els in the solid-state literature lack the chemical reactions and
their size; the growth of the nanocrystals is monitored by the rate definitions that are necessary to clearly define kinetically
shift in the absorbance to higher wavelength. The sizes of the and mechanistically the system under study. Instead, theories
Table B.1
Kinetic models for solid-state reactions
Date Author(s) Reaction system Rate equation(s) Comments Ref.
− π3 Nv G3 t 4
1939 Johnson and None, but use model f (t) = 1 − e ; Equation becomes more complex when [145]
Mehl to fit various existing f (t) = fraction of reactant transformed to product nucleation and growth rates are allowed
experimental data Nv = nucleation rate, to change during the reaction;
G = growth rate, Good fits to experimental data
t = time
1940 Avrami None; uses general N = N (1 − e−nt ); Lack of chemical equations causes [146]
nucleation theory N = number of growth nuclei, confusion about what the rate equation
N = number of germ nuclei, really describes physically
n = probability of growth nucleus formation
p
1944 Prout and 2KMnO4 → MnO2 pf −p = e
k(t−tmax ) ; Induction period is removed in order to [147]
Tompkins + KMnO4 + O2 p = pressure, get a good fit;
pf = final pressure, Does not fit data for AgMnO4
k = rate constant (different for acceleratory vs deceleratory decomposition;
periods), Two different equations needed for
tmax = time of maximum rate acceleratory and deceleratory parts of
the reaction
n
1946 Erofe’ev None; uses α = 1 − e−kt ; Uses empirically determined rate laws, [148]
generalized theory α = fraction of reactant transformed to product, Mott’s nucleation theory for irreversible
n = “dimensionality” of reaction vs steady-state reaction
1971 Šestàk and None General equation: dα m n
dt = kα (1 − α) (− ln(1 − α)) ;
p Begins to resemble autocatalysis (rate [150]
Berggren m, n, and p are exponents that depend on the type of increases with amount of product);
nucleation and growth Used to describe various modes of
nucleation and growth, with changing
exponents; the exponents are not given
clear physical definitions.
Later (in 1991) [149] found to not be
general (i.e., does not agree with other,
higher-order rate equations)
1975 Ng None General equation: dα
dt = kα
1−p (1 − α)1−q ; Gives analysis of the given equation in [151]
p and q are exponents between 0 and 1, not clearly defined relation to Avrami, Prout–Tompkins,
Erofe’ev, and Roginskii–Shulz
equations; assert that these equations
differ only in their values of p and q
1984 Cardew et None, but use their α= t4

3 )/2 ;
(τN τG
A more general model that reduces to [152]
al. model to fit NH4 NO3 Avrami equation at early times;
τN = nucleation time,
decomposition data Introduces many complicating factors,
τG = growth time
reduced variables that lose chemical
meaning;
Shows a dependence of size on ratio of
growth to nucleation rates
1996 Burnham et Coal and kerogen dα = kα(1 − 0.99α)m Similar to Avrami equation except for [153]
dt
al. pyrolysis and the introduction of the 0.99 term to
maturation account for non-zero initial rate
a − ln a0 = k(t − t );
1997 Jacobs None, but uses model ln 0 Called the “generalized Prout–Tompkins [154]
1− 2aa a
1− 0
to fit Prout and i 2ai equation”;
Tompkins’ AgMnO4 a = fraction of reactant transformed to product, Fits the AgMnO4 data better than
data ai = value of a at the inflection point, Prout–Tompkins reaction, in which the
a0 = value of a at the end of the induction period, kinetic curve is not symmetric about the
t0 = induction period inflection point;
Claims that the induction period is due
to slow nucleus growth
2004 Skrdla None, but uses model dα = k(1 − α) + k (1 − α) k + α

; Used to fit data from Prout and [155]
dt k
to fit Prout and k = nucleation rate constant, Tompkins;
Tompkins’ AgMnO4 k = growth (“nucleus branching”) rate constant Still does not fit the induction period
data well
(continued on next page)
Table B.1 (continued)

Date Author(s) Reaction system Rate equation(s) Comments Ref.
dx = e− αt eβt −1 − 2αβ eβt 2 −1 + α eβt 2 −1
;
2
2005 Skrdla and None, but uses model dt t Starts from ideal monatomic gas [156]
t2
Robertson to fit Prout and for “acceleratory kinetics”; assumption, Maxwell–Boltzmann
2
Tompkins’ AgMnO4 dx = eα te−β t −1 (−2α β te−β t 2 −1 + α e−β t 2 −1 ) distribution of energies
data dt
for “deceleratory kinetics” ;
α, β, α , and β are constants containing rate constant terms
and various physical constants
are developed and then either left as is or modified to fit prior

literature data.
The most significant problem with the theories described
above is that they do not fit all of the experimental data. In
fact, some of the models only do well from approximately 1/4
to at most 3/4 of the total reaction when only one model is
used [155,156]. The reason for this is that most of the models
do not account for the induction period seen in the decom-
position data. To deal with the induction period, the models
either shift the data so that the initial time is at the end of the
induction period (by adding a (t − t0 ) term where t0 is the in-
duction time), or the models use different equations for different
parts of the experimental data. Even the more complete model
put forth by Skrdla in 2004 (entry 9 in Table B.1) cannot fit
well the induction period of the classic Prout–Tompkins data Fig. B.1. Prout and Tompkins’ data for the decomposition of AgMnO4 . Adapted
for AgMnO4 decomposition, Fig. B.1 [155]. A modification of from Ref. [147b].
the model was developed the following year (entry 10 in Ta-
ble B.1), and could fit the induction period much better [156]. This autocatalysis is also seen in the rate equations; for ex-
However, this model relies on even more added parameters that ample, the Šestàk–Berggren equation (entry 4 in Table B.1)
lack significant chemical meaning. Adding to the confusion of with p = 0 has the rate equation dα/dt = kα m (1 − α)n , which,
this model is that it employs kinetic energy distributions based assuming an elementary reaction, represents the reaction nA +
on a monatomic ideal gas in order to model the kinetics of solid- mB → (n + m)B, with [A] = (1 − α) and [B] = α. Studies are
state reactions. currently in progress to determine if the Finke–Watzky 2-step
Recently, the simpler, more chemically precise and thus in- nanocluster formation mechanism is indeed related to the solid-
tuitive Finke–Watzky 2-step model, A → B, A + B → 2B, has state kinetics described above [157].
been used to fit solid-state kinetic data [157]. Note that a single
equation is used for the entire reaction—that is, the induction Appendix C. Protein agglomeration
period is treated explicitly and simply. In addition, the Finke–
Watzky model uses separate rate constants for nucleation and Mechanistic studies (theoretical and experimental) of the ag-
growth, and all of the parameters therein (see Eq. (8) in the glomeration of proteins have been underway since the early
main text) have clear physical meanings—rate constants and 1960s [158]. Nucleation theory has been used to describe
initial complex concentrations are the only variables needed to the agglomeration (or polymerization) of proteins. The ag-
describe the system. glomeration of proteins has been implicated in diseases such
Close examination of the solid-state kinetics literature and as Alzheimer’s disease [159], Huntington’s disease [79,160],
related equations (Table B.1) suggests the hypothesis that the Parkinson’s disease [161] and transmissible spongiform en-
Finke–Watzky mechanism is a simpler, more chemical version cephalopathies [162], the most infamous of the latter being
of the more complex solid-state kinetic models. For example, bovine spongiform encephalopathy (mad-cow disease). Analy-
the classic Prout–Tompkins system of decomposing AgMnO4 ses of several different protein agglomeration mechanisms have
[147] can be expressed as two steps, Eq. (B.2)—a nucleation suggested that the process is autocatalytic, albeit typically with-
and then autocatalytic growth mechanism. This is the Finke– out quantitative kinetic fits demonstrating autocatalysis, A +
Watzky mechanism: A → B, A + B → 2B. B → 2B [162].
As early as the 1980s it was observed that “homogeneous
AgMnO4 AgMnO3 + 1/2O2 , (B.2a) nucleation is an essential part of the kinetic mechanism for
AgMnO4 + AgMnO3 2AgMnO3 + 1/2O2 , (B.2b) sickle hemoglobin polymerization” [163]. For this system of
hemoglobin polymerization, a “double nucleation” mechanism
was put forward, which is described in the following way. Ho-
2AgMnO4 2AgMnO3 + O2 . (B.2c) mogeneous nucleation occurs until a critical nucleus is attained.
Fig. C.1. Concentration profile of the aggregated polyglutamine as monitored

by circular dichroism (dark squares), high-performance liquid chromatography
(dark circles), light scattering (open squares), and thioflavin T (open circles). Fig. C.2. Fit of the data for peptide aggregation shown in Fig. C.1 to
The fit to the data (line) is to Eq. (C.1) below. Reprinted with permission from the 2-step Finke–Watzky mechanism in Scheme 4 of the main text. The
Ref. [79]. data was converted to % non-aggregated protein and was fit with the an-
alytic equation (8) with resultant rate constants: k1 = 2.2(6) × 10−3 h−1 ,
k2 = 3.6(6) × 10−4 M−1 h−1 .
This critical nucleus then grows to a long polymer chain. A new
polymer chain can form either by the same homogeneous nu-
cleation mechanism or by heterogeneous nucleation, in which
monomers attach to an existing polymer chain and then grow
into a new chain from there. The double nucleation mechanism
was found to fit experimental data very well, as long as the
nucleus size was allowed to change with respect to monomer
concentration [162].
In 2002, Ferrone et al. reported a study of the kinetics of
polyglutamine aggregation and its effect of the onset age of
Huntington’s disease [79]. The critical nucleus in this system
was found to be one monomer; this suggests that the folding of
a single monomer is the rate-limiting step, after which growth
of the misfolded protein oligomer is autocatalytic and fast. Per- Fig. C.3. Proposed mechanism for polyglutamine aggregation consisting of (a)
haps the most significant and interesting finding from this ki- compacting of a chain (which is the formation of a critical nucleus), (b) addition
netic study is that the energy difference between benign and of another chain to the compacted structure, (c) compacting of the added chain,
pathological polyglutamine is less than 1 kcal/mol; that is, the and (d) repeat of the process. Reprinted with permission from Ref. [79].
implication is that processes differing by less than 1 kcal/mol
might conceivably be involved in Huntington’s disease! The vs log(protein concentration) gives a linear plot with a slope
knowledge of the nucleation and autocatalytic growth mecha- equal to n∗ + 2; that is, such a plot can give the (kinetic) critical
nism of protein agglomeration is obviously extremely important nucleus size [79].
for finding ways to stop the onset of this serious disease. It was The data in Fig. C.1 bears a striking resemblance to the 1997
noted in a 2001 Nature Hypothesis that for Huntington’s and transition-metal nanocluster formation mechanism in Scheme 4
related diseases, “therapy should be aimed at preventing or re- in the main text [62]. In fact, the data are reasonably well fit by
versing aggregation” [160]. We would say that the identification the 2-step Finke–Watzky mechanism, Fig. C.2, an insight not
of the nucleation process of the aggregates needs to be one fo- previously published. In addition, the pictorial mechanism pro-
cus of disease prevention. posed by Ferrone et al. for the agglomeration of polyglutamine,
The aggregation of polyglutamine in Ferrone’s study was Fig. C.3 resembles that of the formation of transition-metal
monitored by circular dichroism (CD), HPLC, light scattering, nanoclusters shown in Scheme 4 in the main text. Fitting the
and thioflavin T methods [79]. The results are shown in Fig. C.1 polyglutamine agglomeration data to the Finke–Watzky 2-step
above. The data are fit by the empirical equation (C.1), where mechanism, with A as the monomer and B as the polymer (and
is the concentration of monomer that has gone to polymer, possibly C as larger, aggregated polymers) has the potential to
simplify, and perhaps at the same time provide additional in-
1 2
= k+ Kn∗ c3 t 2 , (C.1) sights into, the mechanistic studies of protein agglomeration, so
2 that further studies are in progress and will be reported shortly
k+ is the forward elongation rate constant, Kn∗ is the equilib- [137]. Overall, it appears that nucleation and growth mecha-
rium constant for the nucleation of monomers, c is the bulk con- nisms in Nature share key features in common between at least
centration of monomer, and t is time. One of the more interest- nanocluster formation, solid-state transitions, and protein ag-
ing findings of this paper is that plotting log(agglomeration rate) glomeration, to cite examples presented in this review.
References [38] B.J. Berne, R. Pecora, Dynamic Light Scattering with Applications to
Chemistry, Biology, and Physics, Dover Publications, New York, 2000.
[1] D. Kashchiev, Nucleation: Basic Theory with Applications, Butterworth [39] W. Brown (Ed.), Dynamic Light Scattering: The Method and Some Ap-
Heinmann, Oxford, 2000. plications, Oxford Univ. Press, Oxford, 1993.
[2] (a) R. McGraw, J. Chem. Phys. 75 (1981) 5514; [40] (a) C. Sangregorio, M. Galeotti, U. Bardi, P. Baglioni, Langmuir 12
(b) V. Ruth, J.P. Hirth, G.M. Pound, J. Chem. Phys. 88 (1988) 7079; (1996) 5800;
(c) S.L. Girshick, C.-P. Chiu, J. Chem. Phys. 93 (1993) 1273; (b) C. Graf, A. van Blaaderen, Langmuir 18 (2002) 524;
(d) A. Dillmann, G.E.A. Meier, J. Chem. Phys. 94 (1991) 3872; (c) J. Raula, J. Shan, M. Nuopponen, A. Niskanen, H. Jiang, E.I. Kaup-
(e) X.C. Zeng, D.W. Oxtoby, J. Chem. Phys. 94 (1991) 4472; pinen, H. Tenhu, Langmuir 19 (2003) 3499;
(f) Y. Viisanen, R. Strey, H. Reiss, J. Phys. Chem. 99 (1993) 4680; (d) S.G. Thoma, A. Sanchez, P.P. Provencio, B.L. Abrams, J.P. Wilcoxon,
(g) V.B. Fenelonov, G.G. Kodenyov, V.G. Kostrovsky, J. Phys. Chem. J. Am. Chem. Soc. 127 (2005) 7611.
B 105 (2001) 1050. [41] J.A. Creighton, D.G. Eadon, J. Chem. Soc. Faraday Trans. 87 (1991)
[3] (a) J. Huang, L.S. Bartell, J. Phys. Chem. 99 (1995) 3924; 3881.
(b) I.J. Ford, J. Phys. Chem. B 105 (2001) 11649; [42] A. Henglein, M. Giersig, J. Phys. Chem. B 104 (2000) 6767.
(c) D.B. Dickens, J.J. Sloan, J. Phys. Chem. A 106 (2002) 10543; [43] J. Wang, H.F.M. Boelens, M.B. Thathagar, G. Rothenberg, Phys. Chem.
(d) G.W. Turner, L.S. Bartell, J. Phys. Chem. A 109 (2005) 6877. Chem. Phys. 5 (2004) 93.
[4] (a) J.N. Watson, L.E. Iton, R.I. Keir, J.C. Thomas, T.L. Dowling, J.W. [44] A.V. Gaikwad, G. Rothenberg, Phys. Chem. Chem. Phys. 8 (2006)
White, J. Phys. Chem. B 101 (1997) 10094; 1.
(b) K. Liang, G. White, D. Wilkinson, L.J. Ford, K.J. Roberts, W.M.L. [45] U. Kaiser, D.A. Muller, J.L. Grazul, A. Chuvilin, M. Kawasaki, Nat.
Wood, Cryst. Growth Des. 4 (2004) 1039; Mater. 1 (2002) 102.
(c) E. Lyall, P. Mougin, D. Wilkinson, K.J. Roberts, Ind. Eng. Chem. [46] C.A. Jaska, I. Manners, J. Am. Chem. Soc. 126 (2004) 9776.
Res. 43 (2004) 4947. [47] C.M. Hagen, J.A. Widegren, P.M. Maitlis, R.G. Finke, J. Am. Chem.
[5] L. Qu, W. Yu, X. Peng, Nano Lett. 4 (2004) 465. Soc. 127 (2005) 4423.
[6] G. Schmid (Ed.), Clusters and Colloids: From Theory to Applications, [48] G. Schmid, Chem. Rev. 92 (1992) 1709.
VHC, New York, 1994. [49] D.B. Williams, C.B. Carter, Transmission Electron Microscopy, Plenum
[7] R.G. Finke, in: D.L. Feldheim, C.A. Foss Jr. (Eds.), Metal Nanoparti- Press, New York, 1996, Chapter 4.
cles: Synthesis, Characterization, and Applications, Marcel Dekker, New [50] Q. Wu, W. Chen, T.E. Madey, J. Phys. Chem. B 106 (2002) 6419.
York, 2001, Chapter 2. [51] Q. Wu, T.E. Madey, Surf. Sci. 555 (2004) 167.
[52] B.K. Min, W.T. Wallace, A.K. Santra, D.W. Goodman, J. Phys. Chem.
[8] M.A. Watzky, R.G. Finke, Chem. Mater. 9 (1997) 3083.
B 108 (2004) 16339.
[9] D.W. Oxtoby, Acc. Chem. Res. 31 (1998) 91.
[53] In situ (so-called “environmental”) STM measurements of metal surfaces
[10] (a) D. Kashchiev, J. Phys. Chem. 76 (1992) 5098;
are more common; for example: V. Maurice, H.-H. Strehblow, P. Marcus,
(b) D.W. Oxtoby, J. Phys. Condens. Matter 4 (1992) 7627;
Surf. Sci. 458 (2000) 185.
(c) A. Laaksonen, V. Talanquer, D.W. Oxtoby, Annu. Rev. Phys.
[54] G. Margaritondo, Elements of Synchrotron Light: For Biology, Chem-
Chem. 46 (1995) 489.
istry, and Medical Research, Oxford Univ. Press, Oxford, 2002.
[11] D.H. Everett, Basic Principles of Colloid Science, Royal Society of
[55] A.I. Frenkel, C.W. Hills, R.G. Nuzzo, J. Phys. Chem. B 105 (2001)
Chemistry, London, 1988.
12689.
[12] S.D. Lubetkin, Langmuir 19 (2003) 2575.
[56] M.S. Nashner, A.I. Frenkel, D. Somerville, C.W. Hills, J.R. Shapley, R.G.
[13] F. Ferrone, Methods Enzymol. 309 (1999) 256.
Nuzzo, J. Am. Chem. Soc. 120 (1998) 8093.
[14] N.M. Dixit, C.F. Zukoski, Phys. Rev. E 66 (2002) 051602.
[57] A.I. Frenkel, J. Synchrotron Rad. 6 (1999) 293.
[15] M. Volmer, A. Weber, Z. Phys. Chem. (Leipzig) 119 (1926) 227.
[58] L. Menard, A.I. Frenkel, R.G. Nuzzo, E.E. Finney, C. Graham, W.M.
[16] M. Volmer, Kinetik der Phasenbildung, Steinfopff, Leipzig, 1939.
Alley, R.G. Finke, experiments in progress.
[17] R. Becker, W. Döring, Ann. Phys. 24 (1935) 719.
[59] V.K. LaMer, R.H. Dinegar, J. Am. Chem. Soc. 72 (1950) 4847.
[18] A. Dillmann, G.E.A. Meier, Chem. Phys. Lett. 160 (1989) 71. [60] V.K. LaMer, Ind. Eng. Chem. 44 (1952) 1270.
[19] S. Auer, D. Frenkel, Ann. Rev. Phys. Chem. 55 (2004) 333. [61] E. Matijevic, Chem. Mater. 5 (1993) 412.
[20] D.M. Martínez, F.T. Ferguson, R.H. Heist, J.A. Nuth III, J. Chem. Phys. [62] M.A. Watzky, R.G. Finke, J. Am. Chem. Soc. 118 (1997) 10382.
115 (2001) 310. [63] T. Sugimoto, F. Shiba, T. Sekiguchi, H. Itoh, Colloids Surf. A 164 (2000)
[21] B.N. Hale, Phys. Rev. A 33 (1986) 4156. 183.
[22] L. Grànàsy, J. Phys. Chem. 100 (1996) 10768. [64] T. Sugimoto, F. Shiba, Colloids Surf. A 164 (2000) 205.
[23] Y.G. Chushak, L.S. Bartell, J. Phys. Chem. B 105 (2001) 11605. [65] J. Turkevich, P.C. Stevenson, J. Hillier, Faraday Discuss. Chem. Soc. 11
[24] M. Rusyniak, V. Abdelsayed, J. Campbell, M.S. El-Shall, J. Phys. Chem. (1951) 55.
B 105 (2001) 11866. [66] C. Besson, E.E. Finney, R.G. Finke, Chem. Mater. 17 (2005) 4925.
[25] M. Rusyniak, M.S. El-Shall, J. Phys. Chem. B 105 (2001) 11873. [67] E.E. Finney, L.S. Ott, R.G. Finke, manuscript in preparation.
[26] R. Strey, P.E. Wagner, Y. Viisanen, J. Phys. Chem. 98 (1994) 7748. [68] A.F. Schmidt, V.V. Smirnov, Top. Catal. 32 (2005) 71.
[27] T. Sugimoto, J. Colloid Interface Sci. 181 (1996) 259. [69] J.A. Widegren, J.D. Aiken III, S. Özkar, R.G. Finke, Chem. Mater. 13
[28] T. Sugimoto, J. Colloid Interface Sci. 183 (1996) 299. (2001) 312.
[29] T. Sugimoto, F. Shiba, J. Phys. Chem. B 103 (1999) 3607. [70] E.E. Finney, R.G. Finke, unpublished result.
[30] U. Gasser, E.R. Weeks, A. Schofield, P.N. Pusey, D.A. Weitz, Sci- [71] J.A. Widegren, M.A. Bennett, R.G. Finke, J. Am. Chem. Soc. 125 (2003)
ence 292 (2001) 258. 10301.
[31] P.R. ten Wolde, M.J. Ruiz-Montero, D. Frenkel, Phys. Rev. Lett. 75 [72] V. de Vos, Science 279 (1998) 1710.
(1995) 2714. [73] Y. Lin, R.G. Finke, Inorg. Chem. 33 (1994) 4891.
[32] W. Ostwald, Z. Phys. Chem. 22 (1897) 289. [74] B.J. Hornstein, R.G. Finke, Chem. Mater. 16 (2004) 139.
[33] Y.C. Shen, D.W. Oxtoby, Phys. Rev. Lett. 77 (1996) 3585. [75] C. Besson, E.E. Finney, R.G. Finke, J. Am. Chem. Soc. 127 (2005) 8179.
[34] M.S. Nashner, A.I. Frenkel, D.L. Adler, J.R. Shapley, R.G. Nuzzo, J. [76] I.R. Epstein, Nature 374 (1995) 321.
Am. Chem. Soc. 119 (1997) 760. [77] I. Nagypál, I.R. Epstein, J. Phys. Chem. 90 (1986).
[35] A. Cacciuto, D. Frenkel, J. Phys. Chem. B 109 (2005) 6587. [78] E.E. Finney, R.G. Finke, manuscript in preparation.
[36] A.E. Nielson, Kinetics of Precipitation, Pergamon Press, Oxford, 1964. [79] S. Chen, F.A. Ferrone, R. Wetzel, Proc. Natl. Acad. Sci. 99 (2002) 11884.
[37] A.E. Nielson, Kristal Tech. 4 (1969) 17. [80] Y. Lin, R.G. Finke, J. Am. Chem. Soc. 116 (1994) 8335.
[81] A. Cacciuto, S. Auer, D. Frenkel, Nature 428 (2004) 404. (d) A.D. Logan, K. Sharoudi, A.K. Datye, J. Phys. Chem. 95 (1991)
[82] P. Pertuci, G. Vituli, Inorg. Synth. 22 (1983) 178. 5568.
[83] L.E. Crascall, J.L. Spencer, Inorg. Synth. 28 (1990) 126. [117] H. Bönnemann, G. Braun, W. Brijoux, R. Brinkmann, A.S. Tilling, K.
[84] K. Moseley, P.M. Maitlis, Chem. Commun. (1971) 982. Seevogel, K. Siepen, J. Organomet. Chem. 520 (1996) 143.
[85] J.R. Platt, Science 146 (1964) 347. [118] J.D. Aiken III, R.G. Finke, J. Am. Chem. Soc. 121 (1999) 8803.
[86] M. Michaelis, A. Henglein, J. Phys. Chem. 96 (1992) 4719. [119] L.S. Ott, R.G. Finke, Coord. Chem. Rev. 251 (2007) 1075.
[87] J. Belloni, Catal. Today 113 (2006) 141. [120] E.E. Finney, A.I. Frenkel, R.G. Finke, experiments in progress.
[88] A. Henglein, R. Tausch-Treml, J. Colloid Interface Sci. 80 (1981) 84. [121] J.B. Michel, J.T. Schwartz, in: B. Delmon, P. Grange, P.A. Jacobs, G.
[89] A. Henglein, M. Giersig, J. Phys. Chem. B 103 (1999) 9533. Poncelet (Eds.), Preparation of Catalysts: Scientific Bases for the Prepa-
[90] K. Angermund, M. Bühl, E. Dinjus, U. Endruschat, F. Gassner, H.-G. ration of Heterogeneous Catalysts, vol. IV, Elsevier, New York, 1987,
Haubold, J. Hormes, G. Köhl, F.T. Mauschick, H. Modrow, R. Mörtel, pp. 669–687.
R. Mynott, B. Tesche, T. Vad, N. Waldöfner, H. Bönnemann, J. Phys. [122] H. Yu, P.C. Gibbons, K.F. Kelton, W.E. Buhro, J. Am. Chem. Soc. 123
Chem. B 41 (2002) 4041. (2001) 9198.
[91] H. Bönnemann, N. Waldöfner, H.-G. Haubold, T. Vad, Chem. Mater. 14 [123] N.R. Jana, L. Gearheart, C.J. Murphy, Chem. Mater. 13 (2001) 2313.
(2002) 1115. [124] (a) T.R. Lee, G.M. Whitesides, Acc. Chem. Res. 25 (1992) 266;
[92] H.-G. Haubold, T. Vad, N. Waldöfner, H. Bönnemann, J. Appl. Cryst. 36 (b) T.M. Miller, A.N. Izumi, Y.-S. Shih, G.M. Whitesides, J. Am. Chem.
(2003) 617. Soc. 110 (1988) 3146.
[93] L. Beuermann, W. Maus-Friedrichs, S. Krischok, V. Kempter, S. Bucher, [125] E.E. Finney, R.G. Finke, Inorg. Chim. Acta 359 (2006) 2879.
H. Modrow, J. Hormes, N. Waldöfner, H. Bönnemann, Appl. Organomet. [126] D. Chen, R. Wang, I. Arachchige, G. Mao, S.L. Brock, J. Am. Chem.
Chem. 17 (2003) 268. Soc. 126 (2004) 16290.
[94] K. Angermund, M. Bühl, U. Endruschat, F.T. Mauschick, R. Mörtel, R. [127] (a) P.J. Hagrman, D. Hagrman, J. Zubieta, Angew. Chem. Int. Ed. 38
Mynott, B. Tesche, N. Waldöfner, H. Bönnemann, G. Köhl, H. Modrow, (1998) 2638;
J. Hormes, E. Dinjus, F. Gassner, H.-G. Haubold, T. Vad, M. Kaupp, J. (b) K. Yamamoto, Y. Sakata, Y. Nohara, Y. Takahashi, T. Tatsumi, Sci-
Phys. Chem. B 107 (2003) 7507. ence 300 (2003) 470;
[95] C.M. Hagen, L. Vieille-Petit, G. Laurenczy, G. Süss-Fink, R.G. Finke, (c) T.M. Davis, T.O. Drews, H. Ramanan, C. He, J. Dong, H. Schnab-
Organometallics 24 (2005) 1819. legger, M.A. Katsoulakis, E. Kokkoli, A.V. McCormick, R.L. Penn, M.
[96] F. Wen, H. Bönnemann, Appl. Organomet. Chem. 19 (2005) 94. Tsapatsis, Nat. Mater. 5 (2006) 400.
[97] S. King, K. Hyunh, R. Tannenbaum, J. Phys. Chem. B 107 (2003) 12097. [128] M.R. Zachariah, M.J. Carrier, J. Aerosol Sci. 30 (1999) 1139.
[98] F. Ungváry, L. Markó, J. Organomet. Chem. 71 (1974) 283. [129] M.R. Zachariah, M.J. Carrier, E. Blaisten-Bajoras, J. Phys. Chem. 100
[99] L.S. Ott, B.J. Hornstein, R.G. Finke, Langmuir 22 (2006) 9357. (1996) 14856.
[100] R. Tannenbaum, Langmuir 13 (1997) 5056. [130] T. Hawa, M.R. Zachariah, Phys. Rev. B 69 (2004) 035417-1.
[101] J. van Embden, P. Mulvaney, Langmuir 21 (2005) 10226. [131] S.H. Ehrman, S.K. Friedlander, M.R. Zachariah, J. Mater. Res. 14 (1999)
[102] E.V. Shevchenko, D.V. Talapin, H. Schnablegger, A. Kornowski, Ö. Fes- 4551.
tin, P. Svedlindh, M. Haase, H. Weller, J. Am. Chem. Soc. 125 (2003) [132] S.-H. Suh, M.R. Zachariah, S.L. Girshick, J. Vac. Sci. Technol. A 19
9090. (2001) 940.
[103] Y. Chen, J.L. Fulton, J.C. Linehan, T. Autrey, J. Am. Chem. Soc. 127 [133] S.-H. Suh, M.R. Zachariah, S.L. Girshick, Aerosol Sci. 33 (2002)
(2005) 3254. 943.
[104] The Hydrogen Economy: NRC and NAE, The National Academies Press, [134] L.C. Ciacchi, W. Pompe, A. De Vita, J. Am. Chem. Soc. 123 (2001) 7371.
Washington, DC, 2004. [135] L.C. Ciacchi, W. Pompe, A. De Vita, J. Phys. Chem. B 107 (2003) 1755.
[105] C.A. Jaska, K. Temple, A.J. Lough, I. Manners, J. Am. Chem. Soc. 125 [136] L.C. Ciacchi, M. Mertig, W. Pompe, S. Meriani, A. De Vita, Platinum
(2003) 9424. Met. Rev. 47 (2003) 98.
[106] J.C. Linehan, personal communication. [137] J.N. Agar, A.M. Morris, R.G. Finke, Biochemistry (2007), in press.
[107] O. Alexeev, B.C. Gates, Top. Catal. 10 (2000) 273. [138] C.B. Murray, D.J. Norris, M.G. Bawendi, J. Am. Chem. Soc. 115 (1993)
[108] D.G. Duff, A.C. Curtis, P.P. Edwards, D.A. Jefferson, B.F.G. Johnson, 8706.
D.E. Logan, J. Chem. Soc. Chem. Commun. (1987) 1264. [139] X. Peng, J. Wickham, A.P. Alivisatos, J. Am. Chem. Soc. 120 (1998)
[109] R. Choukroun, D. de Caro, B. Chaudret, P. Lecante, E. Snoeck, New J. 5343.
Chem. 25 (2001) 525. [140] W.W. Yu, L. Qu, W. Guo, X. Peng, Chem. Mater. 15 (2003) 2854.
[110] (a) M. Kulzick, R.T. Price, E.L. Muetterties, V.W. Day, Organome- [141] V.N. Soloviev, A. Eichhofer, D. Fenske, U. Banin, J. Am. Chem. Soc. 122
tallics 1 (1982) 1256; (2000) 2673.
(b) Z. Duan, M.J. Hampden-Smith, A.P. Sylwester, Chem. Mater. 4 [142] C.R. Bullen, P. Mulvaney, Nano Lett. 4 (2004) 2303.
(1992) 1146. [143] C.H. Bamford, C.F.H. Tipper (Eds.), Reactions in the Solid State, vol. 22,
[111] (a) P. Espinet, P.M. Bailey, P. Piraino, P.M. Maitlis, Inorg. Chem. 18 Elsevier Scientific, Amsterdam, 1980, pp. 41–113.
(1979) 2706; [144] A.K. Burnham, R.L. Braun, Energy Fuels 13 (1999) 1.
(b) J.S. Ricci, T.F. Koetzle, R.J. Goodfellow, P. Espinet, P.M. Maitlis, [145] W.A. Johnson, R.F. Mehl, Trans. AIME 135 (1939) 416.
Inorg. Chem. 23 (1984) 1823. [146] (a) M. Avrami, J. Chem. Phys. 7 (1939) 1103;
[112] T. Autrey, J.C. Linehan, J.L. Fulton, S. Özkar, R.G. Finke, experiments (b) M. Avrami, J. Chem. Phys. 8 (1940) 212.
in progress. [147] (a) E.G. Prout, F.C. Tompkins, Trans. Faraday Soc. 40 (1944) 488;
[113] Stranski, Phys. Ztsch. 36 (1935) 393. (b) E.G. Prout, F.C. Tompkins, Trans. Faraday Soc. 44 (1946) 468.
[114] C.T. Campbell, S.C. Parker, D.E. Starr, Science 298 (2002) 811. [148] B.V. Erofe’ev, Dokl. Akad. Nauk SSSR 52 (1946) 511.
[115] H. Hirai, Y. Nakao, N. Toshima, J. Macromol. Sci. Chem. A 13 (1979) [149] J. Màlek, J.M. Criado, Thermochim. Acta 175 (1991) 305.
727. [150] J. Šestàk, G. Berggren, Thermochim. Acta 3 (1971) 1.
[116] (a) U. Kolb, S.A. Quaiser, M. Winter, M.T. Reetz, Chem. Mater. 8 (1996) [151] W.-L. Ng, Aust. J. Chem. 28 (1975) 1169.
1889; [152] P.T. Cardew, R.J. Davey, A.J. Ruddick, J. Chem. Soc. Faraday Trans. 80
(b) M.T. Reetz, S.A. Quaiser, M. Winter, J.A. Becker, R. Schafer, U. (1984) 659.
Stimming, A. Marmann, R. Vogel, T. Konno, Angew. Chem. Int. Ed. [153] A.K. Burnham, R.L. Braun, T.T. Coburn, E.I. Sandvik, D.J. Curry, B.J.
Engl. 35 (1996) 2092; Schmidt, R.A. Noble, Energy Fuels 10 (1996) 49.
(c) M. Harada, K. Asakura, Y. Ueki, N. Toshima, J. Phys. Chem. 96 [154] P.W.M. Jacobs, J. Phys. Chem. B 101 (1997) 10086.
(1992) 9730; [155] P.J. Skrdla, J. Phys. Chem. A 108 (2004) 6709.
[156] P.J. Skrdla, R.T. Robertson, J. Phys. Chem. B 109 (2005) 10611. (b) L. Narhi, S.J. Wood, S. Steavenson, Y. Jiang, G.M. Wu, D. Anafi,
[157] J.D. Martin, E.E. Finney, R.G. Finke, experiments in progress. S.A. Kaufman, F. Martin, K. Sitney, P. Denis, J.-C. Louis, J. Wypych,
[158] F. Oosawa, M. Kasai, J. Mol. Biol. 4 (1964) 10. A.L. Biere, M. Citron, J. Biol. Chem. 274 (1999) 9843;
[159] J.D. Harper, P.T. Lansbury Jr., Annu. Rev. Biochem. 66 (1997) (c) T.M. Dawson, V.L. Dawson, Science 302 (2003) 819.
385. [162] (a) J. Masel, V.A.A. Jansen, M.A. Nowak, Biophys. Chem. 77 (1999)
[160] M.F. Perutz, A.H. Windle, Nature 412 (2001) 143. 139;
[161] (a) J.Q. Trojanowski, M. Goedert, T. Iwatsubo, V.M.-Y. Lee, Cell Death (b) M. Eigen, Biophys. Chem. 63 (1996) A1.
Differentiation 5 (1998) 832; [163] F.A. Ferrone, J. Hofrichter, W.A. Eaton, J. Mol. Biol. 183 (1985) 611.

Nucleation and Growth Mechanism of Semicoductors

Uploaded by

Copyright:

Available Formats

Nucleation and Growth Mechanism of Semicoductors

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucleation and Growth Mechanism of Semicoductors

Uploaded by

Copyright:

Available Formats

Journal of Colloid and Interface Science 317 (2008) 351–374

Nanocluster nucleation and growth kinetic and mechanistic studies:

Keywords: Nanoclusters; Nucleation; Growth; Kinetics and mechanism

sites on a surface monomer to that of a free monomer, kB is 3. Nanocluster formation in solution

Fig. 2. The LaMer mechanism of nucleation of sulfur. The (theoretical) curve

Scheme 3. The Finke–Watzky 2-step mechanism for transition-metal nano-

4.2. The Finke–Watzky 2-step, slow, continuous nucleation

The next step in the study of transition-metal nanocluster

which nucleation events are occurring—the formation of the

evidence. In this system, the main method of characterizing the

growth, although this is not certain. The necessary rigorous ki-

4.7. Observation of Rh4–6 clusters in dehydrogenation

In 2005, Chen, Fulton, Linehan and Autrey at Pacific North-

[Cp*RhH]2+ 4 [111]. Computational evidence also supports the

A simple experiment to determine if the solvent is the (major)

6. Computational studies of nanocluster nucleation 7. Conclusions

1984 Cardew et None, but use their α= t4

2004 Skrdla None, but uses model dα = k(1 − α) + k (1 − α) k + α

Table B.1 (continued)

are developed and then either left as is or modified to fit prior

Fig. C.1. Concentration profile of the aggregated polyglutamine as monitored

You might also like