### Solubility of Single-wall Carbon Nanotubes

A new solubility mechanism is based on the possibility of formation of SWNT clusters in solution. Aggregation changes SWNT thermodynamic parameters in solution, which displays the phase equilibrium and changes the magnitude of solubility. The thermodynamic approach to the description of SWNT solubility is based on the

*bundlet* model of clusters, which is valid under conditions when the characteristic number of SWNTs in the cluster

*n* ≫ 1. Let us formulate the problem of determining the temperature dependence of SWNT solubility in terms of the possibility of forming clusters of several parallel SWNTs. In a saturated SWNT solution, the magnitudes of the chemical potential per SWNT for dissolved substance and for a crystal are equal, which is in equilibrium with solution. The equality is valid not only for isolated SWNTs in a solution but also for SWNT clusters. According to the bundlet model of clusters the free energy of a cluster in a solution is made up of two parts: the volume part, proportional to the number of SWNTs

*n* in the cluster, and the surface one, proportional to

*n*
^{1/2} [

23–

25]. The model corresponds to the assumption that clusters consisting of

*n* ≫ 1 particles have a cylindrical bundlet shape and permits the Gibbs energy

*G*
_{
n
} for a cluster of size

*n* to be represented as the sum

where parameters

are responsible for the contribution to the Gibbs energy of molecules placed inside the volume and on the surface of a cluster, respectively. The chemical potential μ

_{
n
} of a cluster of size

*n* in a solution is expressed via

where

*T* is the temperature. Having regard to Eq.

1, this results in

where parameters

are expressed in temperature units. In a saturated solution of SWNTs, the cluster-size distribution function is determined via the equilibrium condition linking the clusters of a specified size with a solid phase, which corresponds to the equality between the magnitudes of the chemical potential (per molecule) for molecules incorporated into clusters of any size and into crystal, resulting in the expression for the cluster-size distribution function in a saturated solution:

where parameter

*A* is the equilibrium difference between the energies of interaction of an SWNT with its surroundings in the solid phase and in the cluster volume,

*B*, the similar difference for SWNTs located on the cluster surface,

*g*
_{
n
}, the statistical weight of a cluster of size

*n*, which depends on both temperature and cluster size

*n*. However, we shall neglect these dependences in comparison with the much stronger exponential dependence in Eq.

4. The presented form (4) for the cluster-size distribution function is based on SWNT structural features. An SWNT is a homogeneous surface structure that, unlike planar or elongated molecules, interacts with its surroundings almost irrespective of the orientation about its axis. The large number of similar elements of the SWNT surface makes it possible to represent the interaction energy of this molecule and the solvent molecules, having essentially smaller size, as the product of a specific surface interaction energy by surface area of the molecule. The feature of the SWNT structure may be further used in the description of the interaction between clusters, made up of SWNTs, and the solvent. This is purely surface interaction and, because the interaction energy of SWNTs with one another, both in a cluster and in a solid is low in comparison with the binding energy of C atoms in an SWNT, one can assume that the specific surface energy of interaction of SWNTs with one another and with solvent molecules is not sensitive to the relative orientation of parallel SWNTs in a cluster. Parameters A and B may have any sign. However, the normalization condition for distribution function (4)

requires

*A* > 0. Here

*C* is the solubility in relative units. As

*n* ≫ 1 normalization (5) may be replaced by integral

Here

is the statistical weight of a cluster averaged over the range of

*n* that makes the major contribution to integral (6), and

*C*
_{0}, the SWNT molar fraction. The

*A* *B* and

*C*
_{0} have been taken equal to those for C

_{60} in hexane, toluene and CS

_{2}:

*A* = 320 K,

*B* = 970 K, C

_{0} = 5 × 10

^{−8} (molar fraction) for

*T* > 260 K. A correction has been introduced to take into account the different packing efficiencies between C

_{60} and SWNTs

where η

_{cyl} = π/2(3)

^{1/2} is the packing eficiency of cylinders, and η

_{sph} = π /3(2)

^{1/2}, that of spheres (face-centred cubic, FCC). The trend of SWNTs in solution to form clusters is reflected in the parameters governing their properties. The dependences of the cluster-size distribution function on solution concentration and temperature lead to the dependences of thermodynamic–kinetic parameters characterizing SWNT behaviour. For an unsaturated solution a solid phase is absent, so that the distribution function is determined via equilibrium condition for clusters. Using Eq.

3, one can obtain the distribution function in the unsaturated SWNT solution depending on concentration:

Here parameter λ depending on the concentration of a solution is determined via normalization condition

*C*
_{0} defines the absolute concentration, can be found by requiring that determined via Eq.

9 to be saturated (Eq. 5) and is taken as 10

^{−4} mol L

^{−1}. The formation energy of a cluster consisting of

*n* SWNTs is determined by

Using the expression for the cluster-size distribution function, one obtains the formula governing the thermal effect of SWNT solution per mole of dissolved substance:

where λ is determined by the total concentration of formed solution via normalization condition (Eq. 9).

### Transfer Phenomena in Single-wall Carbon Nanotube Solutions

The diffusion coefficient is a parameter characterizing the behaviour of fullerenes and SWNTs in solution, which governs their optimum conditions of crystallization, separation and purification. Their diffusion coefficients have a simple estimate in Stokes formula describing the diffusion of a spherical particle in a viscous fluid:

Here

*k* is Boltzmann constant,

*T*, the temperature of the liquid, η, the dynamic viscosity coefficient, and

*r*
_{
s
}, the particle radius. The validity of the equation can be reduced to the requirement of low Reynolds number for a diffusing particle:

where

is the particle characteristic velocity,

*m*, its mass, and ρ, the solvent mass density. Using the relation between the mass of a particle and its radius, the expression provides the minimum radius of a diffusing particle

where ρ

_{
p
} is the particle mass density. Using the characteristic viscosity coefficients of typical organic solvents η∼(1−3) × 10

^{−3} N s m

^{−2}, one obtains that limitation (13a) is reduced to

*r*
_{
s
} ≫ 10

^{−12} m, which is valid for practical purposes. Radii

*r*
_{
s
}, determined by Eq.

12 from experimental data for the diffusion coefficient of fullerenes in various solvents, substantially exceed the radius of a C

_{60} molecule

*r*
_{
s
} = 0.35 nm. The differences in the radii obtained for different solvents may be attributed to fullerene-SWNT aggregation in solution; the effect is universal. The existence of these systems in solution in the form of clusters, whose average size depends on the concentration of solution, suggests the dependence of their diffusion coefficient on concentration [

26]. For low concentration almost no clusters are formed, and their diffusion coefficient is close to the value for a fullerene or SWNT. As the concentration of fullerenes in solution rises, the average cluster size increases and their diffusion coefficient decreases in accordance with Eq.

12. For SWNTs in solution the cluster-size distribution function for saturation is expressed via Eq.

4, whereas for an unsaturated solution, via Eq.

8. Let us determine SWNT diffusion coefficient

*D* in solution based on

Here

*J* is the flux of matter in solution under the action of concentration gradient. In view of the cluster origin of SWNT solubility one represents Eq.

14:

where

*J*
_{
n
}
*D*
_{
n
} and

*C*
_{
n
} are the partial values of the flux, diffusion coefficient and concentration of the cluster of size

*n*, respectively. We shall derive the relationship between diffusion coefficient

*D*
_{
n
} of the cluster of size

*n* and its radius

*r*
_{
n
}, based on the bundlet model, Stokes Eq.

12 and relations

where

*M* is the fullerene molecular mass, and ρ, the cluster density. By combining Eqs. 1416 and using the cluster-size distribution function (8), one derives the expression for the SWNT diffusion coefficient for cluster formation:

Here

*D*
_{0} is the diffusion coefficient of an SWNT. Parameter

*D*
_{0} has been taken equal to that for C

_{60} in toluene:

*D*
_{0} = 10

^{−9} m

^{2}· s

^{−1} at

*T*
_{o} = 295.15 K corrected as

for

*T* ∼

*T*
_{o}. The concentration dependence of the cluster-size distribution function points to a concentration dependence of SWNT diffusion coefficient, which complicates its kinetic behaviour. If a solution contains a mixture of different sorts of SWNTs, the character of the diffusion of SWNTs of a given sort is determined by their propensity to cluster formation. The SWNTs comprising a small admixture to the basic substance do not practically form clusters and are characterized by the diffusion coefficient, which is inherent to SWNT units. The SWNTs of basic substance whose concentration is close to saturated have a trend to aggregation. The diffusion coefficient for this substance exceeds that for an SWNT unit and exhibits the decreasing temperature dependence. The difference in the diffusion coefficients of SWNTs of different sorts makes thinking of developing the diffusion methods of SWNT enrichment, separation and purification. The SWNT that is present in solution as a minor impurity and does not form clusters must have a higher diffusion coefficient than that SWNT whose concentration is close to saturated and that is present in the form of large clusters. We shall assume that the source of SWNTs is provided by a plane layer of a solid material constituting the mixture of SWNTs of two sorts, in which SWNTs of a certain sort predominate whereas the molecules of the other sort make up only a minor impurity [

27]. One can assume that the molecules of minor impurity form almost no clusters and are characterized by SWNT diffusion coefficient

*D*
_{0}. The diffusion coefficient of SWNTs of the predominating sort depends on concentration and, due to the possibility of forming clusters in solution, is lesser than that of isolated SWNTs. The diffusion equations for SWNTs of the predominating sort (concentration

*C*
_{1}) and of the minor impurity (

*C*
_{2}) have the standard form

Here

*D*
_{1} and

*D*
_{2} denote the diffusion coefficients for the first and second components, respectively. Equations 1819 have automodelling solutions dependent on the single variable

*x*/

*t*
^{1/2}; however, for the concentration dependence of the diffusion coefficient the solution calls for numerical calculations. Equation

18 was solved with the initial conditions

which correspond to one-dimensional (1D) diffusion from an instantaneously actuated plane source. Here

is the saturated concentration of SWNTs in solution. The solution of Eq.

19 with the initial conditions

is known quite well at

:

where

*K* is the normalization factor. The solutions to Eqs. 1819 were reported in the form of spatial dependences of SWNT enrichment factor η defined as

We have neglected the difference between the diffusion coefficients of isolated SWNTs of different sorts, which is due to size variation. The enrichment factor of SWNTs some time-dependent distance

*x*
^{*} away from the source assumes the maximum η

_{
m
}. Due to the automodelling character of the solutions of Eqs. 1819 η

_{
m
} is time independent and

The obtained results permit imagining the possible schemes of SWNT diffusion enrichment in solution. It appears appropriate the experience accumulated in the development of isotope separation. We shall consider nonstationary diffusion. A container filled with a solvent is divided into two parts, with a porous partition that does not retard the diffusion motion of dissolved molecules, but prevents convective stirring of the solution in two parts. A SWNT solid mixture with a minor impurity of higher SWNTs is placed at the bottom of one of the parts. Due to the difference in the diffusion coefficients of SWNTs of different sorts, the SWNT mixture penetrating into the second part of the container must be highly enriched with the minor impurity. After a lapse of time corresponding to the maximum value of the enrichment factor for the given system geometry, the second part of the container filled with the enriched solution rapidly drains. The SWNT extract is enriched with the minor impurity in a single-action mode. The diffusion scheme of SWNT enrichment is more convenient in the stationary mode. An elementary separation cell consists of two volumes divided by a porous partition. An initial solution containing SWNTs of two sorts is slowly pumped via one part of the cell. A pure solvent is pumped in the opposite direction via the other part of the cell. Because of diffusion via the porous partition, the solution in the second part of the cell is enriched with the minor impurity. The maximum enrichment factor corresponds to the ratio between the diffusion coefficients for the two components. Because this ratio is ca. 1.3 a multistage system must be used to attain a more significant enrichment factor. The relationship between the resultant enrichment factor η

_{
f
} and the number

*m* of stages is

where η

_{0} is the enrichment factor for a single cell. The method appears most convenient in the enrichment of a solution containing the mixture of a short SWNT with a minor impurity of larger SWNTs. The temperature–concentration dependences of the cluster-size distribution function show the possibility of a new mechanism of SWNT thermal diffusion in solution. We shall define SWNT thermal diffusion coefficient

*D*
_{
T
} in solution by the relation between the thermal diffusion flux

*J*
_{
T
} and the temperature gradient [

28,

29]

We shall assume that the time required for equilibration of the cluster-size distribution function, defined by Eqs. (4–8), is much lesser than that required for smoothing spatial temperature nonuniformities. By Eqs. 4–8 the temperature gradient in solution causes gradients in partial concentrations of clusters, which in turn causes diffusion flows proportional to temperature gradient. The partial diffusion flux of clusters of size

*n* due to temperature gradient is

where the cluster-size distribution function

*f*(

*n*) is given by Eq.

4 or 8, depending on whether the solution is saturated or not. It is assumed that the main temperature dependence of the cluster-size distribution function is in the exponential factor. The net diffusion flux is calculated via the integration of Eq.

25 over

*n*, which permits using Eq.

24 to determine the thermal diffusion coefficient. The diffusion coefficient

*D*
_{
n
} of clusters of size

*n* in solution will be determined again using Stokes Eq.

12, which describes experimental data. The expression for SWNT thermal diffusion coefficient in solution is

The results of calculations, performed for different values of temperature and concentration of the solution of SWNTs in toluene, on the basis of the cluster-size distribution functions (4–8) using Eqs. 12, 25 and 26, showed thermal diffusion, which is a consequence of SWNT aggregation in solution. Only one of the possible mechanisms of SWNT thermal diffusion was treated, which is inherent to fullerenes-SWNTs. Another more general mechanism shows up even in the case of fullerene-SWNT units, which is caused by the larger size of a solute unit as compared with the solvent molecule. For the latter in a temperature gradient, a fullerene molecule is subjected to the action of a force that is proportional to the pressure difference acting from the side of fluid on the two opposing hemispheres of the molecule, which causes a directed drift of molecules whose velocity

*w* may be estimated via Stokes formula

where

*r* is the radius of the fullerene molecule, which results in the estimation of the thermal diffusion coefficient:

Equation 26 differs from the estimate by a factor (−*An* + *Bn*
^{1/2})/*T* ≫ 1. Under conditions favourable to cluster formation the thermal diffusion mechanism, associated with SWNT aggregation in solution, proves much more efficient as compared with the more general mechanism.

### Fractal Structures in Single-wall Carbon Nanotube Solutions

The trend to aggregation of fullerenes-SWNTs in solution manifests in the formation of clusters. Experimental data show that in parallel with small-sized clusters, which form practically in a moment in these solutions, it is possible the formation of large-sized clusters, growing during several months and containing up to several hundred thousands of units. The large cluster growth kinetics in solution was experimentally studied in detail. A solution of C

_{60} in benzene at concentration

which is several times lower than the saturated magnitude, was studied at room temperature using static (SLS) and dynamic light scattering (DLS). The SLS provides the correlation between the relative variation of radiation intensity scattered at a given angle, due to the existence of dissolved matter in solution, and the average mass of particles in this matter, providing the determination of the average mass of fullerene-SWNT clusters. The DLS consists in measuring the spectral line width of scattered radiation due to the Brownian motion (BM) of particles in solution. Because the characteristic velocity of particle BM is inversely proportional to the mean particle radius, this permits the derivation of information on the dimensions of dissolved particles. By combining SLS with DLS one can investigate the dynamics of growth of aggregates in solution, and determine the relation between the mass and size of a cluster. Fullerenes in benzene form fractal aggregates with a fractal dimension ∼2.1. The growth of such structures was observed over a period up to 100 days. The formed structures are unstable and are destroyed by the light shaking of solution, after which the formation and growth of fractal structures is restarted. The growth dynamics of fractal structures gave the measured hydrodynamic radius

*R*
_{
h
} of fractal clusters as a function of time. The behaviour of cluster growth depends on solution preparation. The data correspond to the case when the solution was prepared in the open air. If the solution was prepared in

the measured value of the hydrodynamic radius was ca. 20% higher. The average radius of the fractal cluster at the end of the observation period reaches ∼ 170 nm. In view of the relation between the fractal dimension of a cluster

*D*, its radius

*R* and its number of particles

*n*, i. e.,

where

*r*
_{0} is the radius of the fullerene molecule, one derives that the maximum number of particles in the cluster attained during the observation time of ∼4 × 10

^{6} s is ∼10

^{5}. In a simple model consider an elementary act of coalescence of two particles in a solution under condition (13), when the characteristic value of the Reynolds number for thermal motion of a dissolved molecule is Re ≪ 1 [

30]. The BM can be described in Stokes–Einstein–Smoluchowski approach. Constant

*k* for the aggregation of particles in solution is defined by the diffusion mechanism and expressed by

Here

*r*
_{1} and

*r*
_{2} are the particle radii, and

*D*
_{1} and

*D*
_{2}, their diffusion coefficients in solution. Using Stokes Eq.

12 for particle diffusion coefficient in solution, one derives the rate constant of particle coalescence:

is

for

and

for

*r*
_{1} ≫

*r*
_{2}. The typical value for SWNT saturated concentration in most widely used solvents, corresponding to solubility at room temperature, is

*N*
_{0}∼10

^{18} cm

^{−3}. Their characteristic dynamic viscosity coefficient is η ∼ 0.01 P. The rate constant for coalescence of two SWNTs-clusters of comparable sizes is ∼ 10

^{−12} cm

^{3} s

^{−1}, which corresponds to the characteristic time of the attachment process under diffusion approach

The time required for the equilibrium-size distribution function of small clusters is of the same order. The real time of the growth of fractal clusters (∼ 10

^{6} s) exceeds the estimation result by many orders of magnitude. In describing the growth kinetics of SWNT fractal structures in solution, one must take into account growth mechanism. We shall employ the simple growth models of fractal structures, which are based on the invariability assumption of cluster fractal dimension in its growth process. The simplest model of fractal cluster growth is diffusion-limited cluster aggregation (DLCA). In DLCA cluster aggregation is a result of the attachment of the clusters of comparable sizes. The rate constant is determined from Eqs. 3032 and is virtually independent of cluster size. The growth kinetics of fractal clusters with the average number of particles

*n* is

The right side of Eq.

33 is independent of

*n* because the concentration of clusters of size

*n* is

*N*
_{0}/

*n*, while the attachment of the cluster of size

*n* to the given cluster results in an increase of its size by

*n*. The rate of cluster growth is proportional to the product of both factors and is equal to

*N*
_{0}
*k*. In view of Eq.

29 the DLCA equation of the growth kinetics of a fractal cluster of average size

*n* is

The time required to increase the fractal cluster radius by a factor of 500 is ∼1 s, which differs from the measurement results by six orders of magnitude; DLCA does not apply to experimental conditions. Another model used to describe fractal structure growth is diffusion-limited aggregation (DLA). In DLA cluster growth is the result of attachment to a given cluster of individual particles (SWNTs or small SWNT clusters). If the initial number density

*N*
_{0}of SWNTs in solution and average concentration

*N*
_{
c
}of growing clusters are time-independent, one derives the equation describing the time variation of the average cluster size

*n*:

Here in accordance with Eqs. 29–32 one has

The form of Eq.

35 is independent of the size of a small cluster attaching to a large cluster of size

*n*. Let the number of SWNTs in a small cluster be equal to

*n*
_{
s
}, and the concentration of clusters of this size,

*N*
_{
s
}. The growth rate of large clusters because of the attachment of the small clusters of size

*n*
_{
s
} is written as

The summation of this expression over all values

*s*≪

*n* in view of the obvious normalization condition

provides Eq.

33. The growth rate of large fractal clusters does not depend on the shape of the size distribution function of small clusters. The feature is caused by the form of the cluster size dependence on the attachment rate constant (32), which in the limiting case of clusters of highly differing sizes does not depend on the size of the smaller cluster. The solution of Eq.

33 with the initial condition

*n*(

*t* = 0) = 1 has the form:

Here

is the maximum number of particles in a cluster. Equation

39 is simplified for

*D* = 2:

where

is the maximum cluster radius, and

. In accordance with Eq.

40 the characteristic time of cluster growth is

. The conclusion does not correspond to experiment. Because the dependence

*R*(

*t*) is close to saturation at the last growth stage one may assume that

. The

, and the characteristic time of cluster growth is estimated as

. Because the measured value of this time exceeds the estimation result by nine orders of magnitude, one concludes that DLA is unsuitable for the description of the experimentally examined growth of SWNT fractal clusters in solution. Another model used to describe fractal cluster growth is reaction-limited cluster aggregation (RLCA). In RLCA the cluster growth is a result of the attachment of clusters of different sizes, with the attachment probability of approaching clusters being γ ≪ 1, so that for a pair of clusters to attach they must undergo a large number of collisions. The equation describing the cluster growth kinetics in RLCA is

where

*R*
_{1} and

*R*
_{2} are the radii of approaching clusters, and μ, their reduced mass. Using Eq.

29 and averaging Eq.

41 over the cluster-size distribution function one derives

Here

*r*
_{0} is the fullerene molecular radius, and

*m*
_{0}, its mass. Dimensionless coefficient

*J* depends on the cluster-size distribution function and cluster fractal dimension

*D*. The

*J* = 6.8 for

*D* = 2, and the simplest form of the function,

where

*n*
_{0} is the average number of particles in the cluster. Integration of (42) results

The RLCA leads to an unlimited growth of the cluster radius with time. Because

dependence (44) is close to linear. Such a dependence differs from the experimental curve, which permits concluding that RLCA is not applicable to the growth of fractal SWNT clusters in solution. A satisfactory agreement between the calculated and measured evolution of fractal cluster growth can be reached because of RLCA modification: let us assume that cluster attachment probability γ depend on cluster size

This results in the expression

Equation 46 calculated for *D* = 2.08, α = 2, and γ_{0} = 10^{−7} showed that the dependence agrees quite well with experiment. The calculated dependence almost coincides with the calculation result within the simplified model with *D* = 2.

### Dimensional Analysis for the Early and Later Stages of Fusion-site Expansion

The two stages of cluster fusion, a fast early and a slower later stage, were detected also in vesicle fusion. During the former the fusion site opened rapidly: the expansion velocity of the rim of the site was
. The fusion pore opens up to micrometres within a hundred microseconds. One would relate this time τ_{early} to fast relaxation of membrane tension. The tension of the clusters achieved before fusion was in the stretching regime of the membrane. The τ_{early} should be primarily governed by the relaxation of membrane stretching. Viscous dissipation can be associated with two contributions: in-plane dilatational shear as the fusion site expands and intermonolayer slip among the leaflets of the multilayer membrane in the fusion-site zone. The latter is negligible for fusion-site diameter *L* larger than half a micrometre. The τ_{early} ∼ η_{
s
}/σ, where η_{
s
} is the surface dilatational viscosity coefficient of the membrane
with units [bulk viscosity coeffcient] × [membrane thickness] [32]. For membrane tensions
close to the tension of rupture
one obtains τ_{early} ∼ η_{
s
}/σ ∼ 100μs, in agreement with experiment
. During the later stage of the fusion process the site expansion velocity slowed down by two orders of magnitude. The dynamics was governed by the displacement of volume Δ*V* of fluid around the fusion site between the fused clusters. The restoring force was related to the bending elasticity of the membrane. Decay time τ_{late} ∼ ηΔ*V*/κ where η is the bulk viscosity coefficient of the solvent, Δ*V* ∼ *R*
^{3}, and κ, the bending elasticity modulus of the membrane
. For a cluster size of *R* = 20 μm one obtains τ_{late} ∼ 100 s, which is the time scale measured for complete fusion-site opening. When two clusters fuse at several contact points and form some fusion sites, the coalescence of these fusion sites can lead to small, contact-zone clusters. Consider three fusion sites, which expand and touch each other in such a way that they enclose a roughly triangular segment of the contact zone. If the three sites are circular and have grown up to a diameter *L*
_{1}, the enclosed contact-zone segment will form a contact-zone cluster of radius
, as follows from geometric considerations. The coalescence of these several sites can lead to small contact-zone clusters encapsulating solvent. One expects that these clusters be interconnected by thin tethers, because pinching the membrane off completely would require additional energy. The fusion-induced cluster formation resembles the membrane processes during cell division, when one looks at them in a time-reversed manner. During the initial stages of the division process, the cell accumulates membrane in the form of small vesicles that define the division plane and transform into two adjacent cell membranes. From dimensional analysis it is found an appropriate time scale τ for the later stage of the expansion of the fusion site. The driving force for this expansion is provided by membrane tension σ, whereas the hydrodynamic-Stokes friction is governed by solution viscosity coefficient η. The system is characterized by two well-separated length scales: the membrane thickness *l* and a typical cluster size *R*. It is chosen *R* = (*R*
_{1} + *R*
_{2})/2 where *R*
_{1} and *R*
_{2} are the radii of the two clusters before they were brought into contact. The only time scale, which one can obtain from a combination of the four variables σ, η, *l* and *R*, is given by τ = (η*R*/σ)*f*(*l*/*R*) with the dimensionless function *f*(*l*/*R*). Because *l* ≪ *R* one can replace *f*(*l*/*R*) by *f*(0) and ignore corrections or order (*l*/*R*). Let *v* (in m s^{−1}) be the average site expansion velocity for a single site. The same order of magnitude for the average expansion velocity is deduced if one assumes that the fusion process startes with *N* > 1 fusion sites. The fusion sites would grow until they start to touch and coalesce. They would then create a coalesced site of diameter *L* if each site had grown up to *L*/*N*
^{1/2}, which implies an average expansion velocity
, still of the same order of magnitude even if *N* were as large as 10.