Environ. Sci. Technol.

1996, 30, 1911-1918

Settling Velocities of Fractal coagulation (3) and sedimentation tanks (4) and filtration
columns (5, 6), however, are usually based on the descrip-
Aggregates tion of the particles as spheres. The settling velocities of
aggregates obtained from natural systems (7) or those
produced in the laboratory (8, 9) have also been analyzed
CLIFFORD P. JOHNSON, by assuming that the aggregate has settling properties
XIAOYAN LI, AND BRUCE E. LOGAN* similar or identical to those of impermeable spheres. Such
Department of Chemical and Environmental Engineering, assumptions have made it difficult to reconcile observed
University of Arizona, Tucson, Arizona 85721
and predicted settling velocities of these aggregates.
The settling velocities of isolated impermeable solid
particles are well predicted using widely available equations
Aggregates generated in water and wastewater and correlations (10). Settling velocities of highly porous
treatment systems and those found in natural systems aggregates have also been experimentally measured, al-
are fractal and therefore have different scaling though aggregates in these studies were not shown to be
fractal (11, 12). There have been efforts to calculate settling
properties than assumed in settling velocity calculations
velocities of aggregates as permeable porous spheres (13,
using Stokes’ law. In order to demonstrate that 14) and as permeable fractal aggregates (15), but the
settling velocity models based on impermeable spheres equations used in all these investigations have been based
do not accurately relate aggregate size, porosity on permeability correlations that assume a homogeneous
and settling velocity for highly porous fractal aggregates, distribution of particles within the aggregate. Accurate
we generated fractal aggregates by coagulation of equations for relating settling velocity to other aggregate
latex microspheres in paddle mixers and analyzed properties such as porosity, density, and mass are important
each aggregate individually for its size, porosity, and since these other properties are often calculated from
settling velocity data by assuming that Stokes’ law is valid
settling velocity. Settling velocities of these ag-
(8, 9, 16). The use of inaccurate settling equations could
gregates were on average 4-8.3 times higher than naturally lead to significant errors in these reported
those predicted using either an impermeable sphere aggregate properties.
model (Stokes’ law) or a permeable sphere model that In this study, we present data on the settling velocities
specified aggregate permeability for a homogeneous of a population of fractal aggregates varying in size from
distribution of particles within an aggregate. Fractal 100 to 1000 µm. These aggregates were generated from
dimensions (D) derived from size-porosity relation- dyed latex microspheres in standard paddle mixers and
ships for the three batches of aggregates were 1.78 ( analyzed independently for size, settling velocity, density
0.10, 2.19 ( 0.12 and 2.25 ( 0.10. These fractal and fractal dimension. Our results indicate that fractal
dimensions were used to predict power law relationships aggregates settle on average 4-8.3 times faster (range 2-20
times) than calculated using Stokes’ law even after inclusion
between aggregate size and settling velocity based
of their nonconstant aggregate density. We attribute this
on Stokes’ law. When it was assumed that the the difference to substantially different drag relationships for
drag coefficient, CD, was constant and fixed at its fractal rather than equivalent sized spherical particles. Our
value of CD ) 24/Re for the creeping flow region (Re findings have important implications for the interpretation
, 1), predicted slopes of size and settling velocity of aggregate densities and fractal dimensions determined
were in agreement with only the data sets where D > in previous studies from settling velocity data.
2. As a result, when D < 2, aggregate porosities
will be overestimated and fractal dimensions will be Methods
calculated incorrectly from settling velocity data The settling velocity of an impermeable spherical aggregate
and Stokes’ law. can be predicted from Stokes’ law. Most aggregates
however are not spherical, and it is thought that they are
permeable. This has led to the proposal of several different
scaling relationships between aggregate size and settling
Introduction velocity for fractal aggregates. In order to show how
Particle transport by gravitational sedimentation is im- different assumptions of either Euclidean or fractal proper-
portant in nearly all water and wastewater treatment ties affect predictions of settling velocity as a function of
processes. Particles settle out in clarifiers following chemi- aggregate size, we review Stokes’ law below. This derivation
cal addition and flocculation in conventional water treat- will allow us to show how fractal dimensions have been
ment process trains, and microbial aggregates formed in deduced from size-settling velocity relationships.
activated sludge aeration tanks and other bioreactors are
also removed by settling in clarifiers. Substantial research Theoretical Section. Impermeable Spherical Aggregates.
indicates that these aggregates have fractal geometries (1, The settling velocity of a spherical impermeable aggregate
2). The equations used to model particle settling in is calculated from a force balance, producing Stokes’ law
(10). There are three forces, gravity (Fg), buoyant (Fb), and
* Corresponding author telephone: 520-621-4316; fax: 520-621- drag (Fd), acting upon an aggregate, which balance ac-
6048; e-mail address: [email protected]. cording to

S0013-936X(95)00604-3 CCC: $12.00  1996 American Chemical Society VOL. 30, NO. 6, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1911
 Fg - F b ) F d (1) to the assumption that the aggregate porosity was not
constant. The nonlinear relationship between aggregate
Since Fg ) FaVag, where Fa is the aggregate density, g is the size and porosity is a consequence of the fractal geometry
gravitational constant, and Fb ) F lVag, where Fl is the of aggregates. It is well known for colloidal aggregates that
suspending liquid density, the sum of the gravity and the mass of the aggregate scales with its size, l, according
buoyant forces can be replaced in this relationship as (18) to

Va(Fa - Fl)g ) Fd (2) m ∼ lD (12)

If all particles composing the aggregate have the same where D is a three dimensional fractal dimension. Since
density Fp, the density difference in eq 2 can be equivalently the mass of an aggregate is related to the number of particles
written using the identities in the aggregate by m ) Nmo, it follows that

(Fa - Fl) ) (1 - p)(Fp - Fl) ) (1 - p) ∆F (3) N ∼ lD (13)

where p is the aggregate porosity and ∆F is the difference Using eqs 5, 9, and 13, we can therefore derive the fractal
between the particle and fluid densities. The drag force scaling relationship
exerted on an object is expressed as a function of the fluid
density and the object’s velocity (U), projected area (A), (1 - p) ∼ lD-3 (14)
and an empirical drag coefficient (Cd). Using the common
Assuming that the length scales l and d are the same, the
expression that Fd ) FU2ACd/2 (10), eq 2 can be written as
following scaling relationship can be derived from eqs 8
Va(1 - p)∆Fg ) FlU2ACd/2 (4) and 14 between settling velocity and aggregate size:

At this point in the derivation we must use geometrical U ∼ lD-1 (15)

relationships to simplify eq 4. For spheres Thus it was proposed (1) that the fractal dimension D could
π be derived from the slope of a log-log plot of settling velocity
Va ) d3 (5) and aggregate size.
6
It has been suggested by others (2, 19) that the
π 2 geometrical and empirical relationships (eqs 6 and 7)
A) d (6)
4 assumed to derive eq 12 might not be valid for all fractal
inorganic and organic aggregates. Logan and Wilkinson
24 (2) proposed that the area-length equation also be cast in
Cd ) (Re , 1) (7)
Re terms of a fractal dimension as
where the Reynolds number Re ) Ud/ν, d is the aggregate
A ∼ l D2 (16)
diameter, and ν is the fluid kinematic viscosity (17).
Combining eqs 4-7, produces Stokes’ law: where D2 is a two-dimensional fractal dimension. When
D < 2, a colloidal aggregate viewed in two dimensions is
g∆F(1 - p) 2 transparent since all particles are visible, and D2 ) D.
U) d (8)
18νFl However, when D g 2, then D2 ) 2 (18). When eq 16 is used
instead of eq 6, settling velocity scales according to
For an aggregate made up of N particles each of mass
mo and volume vo, the porosity can be derived in terms of U ∼ lD-D2+1 (17)
Nvo Jiang and Logan (19) proposed a settling velocity
(1 - p) ) (9)
Va relationship that could be used when Reynolds numbers
were not in the creeping flow region. They used drag
where the percent solids in an aggregate is calculated as coefficients developed for spheres only because no such
100(1 - p). Substituting eq 9 into eq 8 and assuming correlations were available for fractal aggregates. They fit
aggregate volume as defined in eq 5 produces the empirical drag correlation (17) with the power law
relationship
g∆FNvo
U) (10)
3πνFld Cd ) aRe-b (18)

The predicted settling velocity in eq 10 can be compared where for Re , 1, a ) 24 and b ) 1 and for 0.1 < Re < 10,
to the actual settling velocity, Uact, using the dimensionless a ) 29.03 and b ) 0.871. Incorporating eqs 16 and 18 into
number, Γ, defined as the derivation of a scaling relationship for settling velocity
produced
Uact
Γ) (11)
U U ∼ l(D-D2+b)/(2-b) (19)
Calculation of a Fractal Dimension from Scaling Rela- These relationships are summarized in Table 1 for different
tionships. It is commonly observed that for aggregates U ranges of Reynolds numbers and fractal dimensions.
is not proportional to d2, but to dc where c is some value Homogeneous Permeable Spherical Aggregates. Flow
less than 2 (7, 8, 16). The reason for this difference from through the interior of an aggregate can increase the settling
Stokes’ law was attributed by Li and Ganczarczyk (1) only velocity of an aggregate compared to otherwise identical

1912 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 6, 1996

TABLE 1
Scaling Relationships Used To Calculate Fractal
Dimensions from Settling Velocity Data Based on
Different Assumptions of Aggregate Properties
general
expression conditions if D < 2 if D g 2 ref
U ∼ lD-1 Re , 1 U ∼ lD-1 U ∼ lD-1 1
U ∼ lD-D2+1 Re , 1 U∼l U ∼ lD-1 2
Re , 1
U ∼ l(D-D2+b)/(2-b) U∼l U ∼ lD-1 19
b)1
0.1 < Re < 10
U ∼ l(D-D2+b)/(2-b) U ∼ l0.77 U ∼ l0.89(D-1) 19
b ) 0.871

but impermeable particles. The settling velocity of a

permeable aggregate was presented by Masumoto and
Suganuma (11) as

Uperm ) U
[ ξ
+
3
ξ - tanh(ξ) 2ξ2 ] (20)

where the dimensionless variable ξ ) d/(2κ1/2) relates

aggregate size to permeability of the porous media. The FIGURE 1. Settling columns used in experiments.
main difficulty in applying eq 20 to fractal aggregates is
that permeability functions used in previous investigations settling aggregates based on calculations using the Faxen
(14, 15) have been developed for homogeneous media. friction factor (22). For spheres 50-500 µm in diameter,
Logan and Hunt (14) used the Davies correlation (12) the drag of the cylinder wall was calculated to increase
drag by less than 3.2%. Settling velocities were measured
1 16 at a depth of 15-20 cm below the release point based on
) (1 - p)1.5[1 + 56(1 - p)3] (21)
κ a 2 experiments indicating that aggregate terminal velocities
cyl
were reached within 15 cm. The top of the column was
where κ is the aggregate permeability and acyl is the radius sealed with a rubber stopper containing a small tube to
of a long filament assumed to form the aggregate. Even allow introduction of aggregates. Sealing the top of the
after accounting for aggregate size-porosity relationships, column was necessary to avoid convective currents in the
Logan and Hunt calculated that the power in a size-settling column during settling experiments.
velocity relationship was only increased slightly from U ∼ The columns were designed with retrieval wells to allow
d0.4 to Uperm ∼ d0.44 as a result of aggregate permeability. the recovery of aggregates after a settling experiment. The
Others have predicted similarly small changes in settling flat bottom of the column base was sealed to the well base
velocities due to aggregate permeability (20). The main using silicon vacuum grease. After a settling experiment,
reason for deviations from Stokes’ law in experimental the column base was pushed to the side and the column
observations (8, 9) is thought to be a result of nonlinear was removed, leaving the aggregate in the bottom well
relationships between aggregate size and porosity (14). (depth ) 0.8 cm) for subsequent analysis.
Experimental Section. Generation of Fractal Aggregates. Settling Experiments. Three separate batches of ag-
The aggregates used in settling experiments were generated gregates generated in the paddle mixers were used in settling
from latex microspheres (Polysciences Inc.) 2.6 µm (red experiments. Water used in the paddle mixer was identical
beads, experiments 1 and 2) or 0.87 µm (yellow-green beads, to that in the settling column. To ensure that the salinity
experiment 3) in diameter with a density of 1.05 g cm-3 of the water in the aggregates was exactly the same as that
specified by the manufacturer and verified by us as 1.050 in the settling column (i.e., to ensure that salinity of the
( 0.003 g cm-3 using a Percoll density gradient. Micro- water in the covered beakers in the paddle mixer did not
spheres (2.5% suspension by weight) were coagulated in change by evaporation), individual aggregates were trans-
NaCl solutions (average density of 2.5%) in a paddle mixer ferred through three separate beakers in series (over a time
(Phipps and Bird Model 7790-400) at 10 rpm using 1000- period of 20 min), each containing water identical to that
mL round beakers filled to 500 mL and standard flat paddles placed in the settling column. Each aggregate was trans-
(19.2 cm2 surface area). A shear rate of 5 s-1 was estimated ferred using a P-1000 pipetteman (Rainin) with the 1-mL
from Lai et al. (21) for these conditions. In experiment 1, pipet tip cut midway to minimize breakup of the aggregate.
aggregates formed within 24 h. Although the same salt The aggregate was then placed into the top of the settling
concentrations were used in all three experiments, ag- column, and the settling velocity at a point 15-20 cm from
gregates generated during experiments 2 and 3 took longer the top of the column was recorded using a Sony CCD
to form (∼48-72 h). This likely resulted in aggregate camera with a macrolens connected to a four-head VCR
breakup and re-aggregation being more important in (JVC Corp.). The column was illuminated using a fiber
forming aggregates in experiments 2 and 3 than these optic light source. The settling velocity was measured using
processes were in experiment 1. a stopwatch as the time for the aggregate to transverse the
Settling Column. Settling experiments were performed field of vision (4.0 ( 0.5 cm) to within 0.25 s.
in 20 cm high acrylic settling columns 3.15 cm in diameter The accuracy of the settling column was tested in settling
mounted on sliding base plates (Figure 1). The diameter experiments using individual polystyrene beads (density
of the column was selected to minimize wall effects on of 1.05 g cm-3) ranging from 400 to 500 µm in diameter

VOL. 30, NO. 6, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1913

FIGURE 2. Comparison of observed (points) settling velocities versus
those predicted using Stokes’ law (lines) for solid spheres (G ) 1.05
g cm-3) in ultrapure water (Gl ) 0.998 g cm-3; ( and solid line) and
a NaCl solution (Gl ) 1.0446 g cm-3; b and dashed line).

(Figure 2). Settling velocities of these beads in ultrapure

water (Milli-Q) of density 0.998 g cm-3 were within 3.5%
of values calculated using Stokes’ law (eq 8). In a NaCl
solution (Fl ) 1.0446 g cm-3), measured settling velocities
were all <10% of predicted values (average 7.2%).
Aggregate Properties. The cross sectional areas of
aggregates were measured by placing the column base
containing an aggregate from a settling experiment on a FIGURE 3. Fractal dimensions of three different batches of aggregates
microscope (Olympus BH-2) stand for aggregate sizing using generated in a paddle mixer calculated from the slope of the
an image analysis system (Galai Scan Array) at 100×. regression lines (A) O, experiment 1, D ) 1.79 ( 0.10 from solid line;
Aggregates were viewed under bright field (red beads) or 4, experiment 2, D ) 2.19 ( 0.12 from dashed line; (B) 0, experiment
blue light (yellow-green fluorescent beads). Aggregate size 3, D ) 2.25 ( 0.10 from solid line).
was calculated as an equivalent diameter using d ) (4A/ beads used to generate the aggregates (2.6 µm in experi-
π)1/2. ments 1 and 2 and 0.87 µm in experiment 3).
After sizing, individual aggregates were collected from Settling Velocities Compared to Impermeable Ag-
the column well using a pipet tip, transferred into a gregates. The observed settling velocities of microsphere
container containing ultrapure water, shaken, sonicated aggregates were consistently higher than those predicted
for 1 h, and filtered onto a plain (red beads) or black (yellow- from Stokes’ Law for aggregates of identical size, mass and
green beads) 0.4 µm pore diameter polycarbonate filter (25 primary particle density (Figure 4). The dimensionless ratio
mm diameter, Poretics Corp.). Filters were mounted onto of the observed to predicted velocities, Γ, ranged from 2 to
glass slides, and 20-50 fields were counted under white or 20 (Figure 5) with averages for experiments 1, 2, and 3 of
blue light at 1000×. The total number of beads was obtained Γ ) 8.3 ( 4.0 (n ) 110), Γ ) 4.0 ( 1.2 (n ) 40), and Γ )
by multiplying the average number of beads per field by 6.5 ( 2.9 (n ) 65), respectively. This ratio did not appear
the ratio of the total and view areas. to vary in proportion to the fractal dimension. All Γ data
sets were significantly different from each other (Mann-
Aggregate encased volume and porosity were calculated
Whitney Rank Sum Test) despite the overlap of the data.
using eqs 5 and 9. The average fractal dimension for each
Levels of significance by experiment number were as
of the three batches of aggregates was calculated using eq
follows: 1 and 2, 2 and 3, p < 0.0001; 1 and 3, p < 0.0041.
13, and the Stokes’ settling velocity was predicted using eq
The predicted slopes of the settling velocity relationships
10. The velocities of permeable aggregates were predicted
were not significantly different than the observed slopes
using Stokes’ law and eqs 20 and 21. Additional details of
for the aggregates with D g 2 and when it was assumed that
the experiments are given in ref 23.
Re , 1 (Table 2, experiments 2 and 3). Since the fractal
dimensions in experiments 2 and 3 were larger than 2, all
Results creeping flow models predicted the slopes to be equal to
Fractal Dimensions. A total of 215 aggregates generated D - 1. Based on the measured fractal dimensions, the
in three separate coagulation experiments were successfully predicted slopes were 1.19 and 1.25, while the observed
captured during settling velocity experiments and analyzed. slopes were 1.20 ( 0.11 and 1.33 ( 0.10. Two different
Aggregates that broke up during settling or any transfer slopes were calculated for the experiment with D < 1
step were discarded. Using eq 13, the fractal dimensions (experiment 1), depending on whether D2 was included
were calculated as 1.79 ( 0.10, 2.19 ( 0.12, and 2.25 ( 0.10 (eq 15) in the calculation or not (eq 17). When D2 was not
for the three different batches of aggregates (Figure 3). included in the scaling relationship, the predicted slope of
Different size-solid volume relationships were produced 0.79 was significantly less than the observed slope of 1.04
in experiments 2 and 3 even though the fractal dimensions ( 0.10. Inclusion of D2 resulted in a predicted slope of 1,
were similar likely as a result of the different sizes of the in agreement with observations.

1914 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 6, 1996

 TABLE 2
Comparison of Slopes from Settling Velocity Data
with Those Predicted by Different Fractal Scaling
Equations
expt 1 expt 2 expt 3
relationship conditions (D < 2) (D g 2) (D g 2)
data Re , 1 1.04 ( 0.10 1.20 ( 0.11 1.33 ( 0.10
U ∼ lD-1 Re , 1 0.79 1.19 1.25
U ∼ lD-D2+1 Re , 1 1 1.19 1.25
Re , 1
U ∼ l(D-D2+b)/(2-b) 1 1.19 1.25
b)1
0.1 < Re < 10
U ∼ l(D-D2+b)/(2-b) 0.77 1.06 1.11
b ) 0.871

FIGURE 4. Settling velocities of aggregates predicted using Stokes’

law (+) versus those observed in experiments: (A) O, experiment
1, D ) 1.79; 4, experiment 2, D ) 2.19; (B) 0, experiment 3, D ) 2.25.
Slopes of the three lines are given in Table 2.

FIGURE 6. Settling velocities of aggregates predicted for permeable

spherical aggregates (×) versus those observed in experiments:
(A) O, experiment 1, D ) 1.79; 4, experiment 2, D ) 2.19; (B) 0,
experiment 3, D ) 2.25. Slopes of the three lines are given in Table
2.

Settling Velocities Compared to Homogeneous Per-

FIGURE 5. Dimensionless ratio (Γ) of observed and predicted meable Aggregates. Settling velocities predicted from
aggregate settling velocities: (A) O, experiment 1, D ) 1.79; 4,
experiment 2, D ) 2.19; (B) 0, experiment 3, D ) 2.25.
permeability relationships developed for aggregates com-
posed of particles distributed homogeneously throughout
the aggregate were appreciably lower than observed
When drag coefficients were calculated based on larger
velocities (Figure 6). The small increases in settling
Re numbers (0.1 < Re < 10) using b ) 0.871 in eq 18, the velocities predicted here are consistent with previous
observed slopes were always significantly larger than those comparisons of impermeable and permeable aggregates
predicted for spherical aggregates using independently (14, 15). The observation that settling velocities of fractal
measured parameters of D and D2 in eq 19 (Table 2). The aggregates are larger than predicted by permeable aggregate
calculated slopes for Re > 0.1, however, included data for models means that permeability relationships derived for
lower Re results (Re < 0.1) when b ) 1. There was a homogeneous distribution of particles in a porous medium
insufficient data to calculate a statistically significant slope (such as eq 21) incorrectly describe the permeabilities of
when data with Re < 0.1 was excluded. We were therefore fractal porous media. The non-homogeneous distribution
unable to test whether there was a significant change in the of particles in fractal aggregates results from the coagulation
slopes for the different Re number ranges of Re < 0.1 and of small and more densely packed clusters into larger and
Re > 0.1. overall less dense aggregates. The permeability of the

VOL. 30, NO. 6, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1915

 the fractal aggregates than spheres results in significantly
lower values of the constant a when D < 3. Although this
comparison suggests that the drag experienced by ag-
gregates with fractal dimensions less than 2 is significantly
different from those with D > 2, we have insufficient data
at this time to include this observation into a method that
incorporates the fractal dimension into a drag coefficient
correlation.

Discussion
These experiments prove that fractal aggregates composed
of inorganic microspheres can settle on average 4-8.3 times
faster (range 2-20) than predicted by calculations for
impermeable or permeable spheres of identical mass, cross
sectional area, and primary particle density. These dif-
ferences in settling velocities are likely a consequence of
the heterogeneous distribution of primary particles in a
fractal aggregate. As fractal aggregates increase in size,
pores become larger, likely permitting greater quantities of
flow through the aggregate interior than possible for
permeable aggregates having a homogeneous distribution
of particles within the aggregate. These large pores produce
a smaller overall drag per total cross sectional area for the
fractal aggregate than calculated for an impermeable or
permeable spherical aggregate.
This finding that aggregates settle faster than predictions
based on modified forms of Stokes’ law is expected from
previous comparisons of simulated settling velocities and
experimentally measured properties of colloidal-sized frac-
FIGURE 7. Drag coefficients calculated as a function of Reynolds
number for a sphere (solid lines) versus experimental data: (A) O, tal aggregates. Simulations of hydrodynamic friction,
experiment 1, D ) 1.79; 4, experiment 2, D ) 2.19; (B) 0, experiment assumed to follow a Stokes-Einstein relationship, over-
3, D ) 2.25. Slopes of the three regression lines (dashed lines) are estimated the friction of fractal objects such as macro-
given in Table 2. molecules. Wiltzius (24) compared the size of aggregates
measured by their radius of gyration, rg, using static light
TABLE 3 scattering, to their hydrodynamic radius, rh, calculated from
Empirical Drag Coefficient (Cd ) aRe-b) Constants quasielastic light-scattering experiments in terms of the
for Different Fractal Dimensions ratio β, defined as

impermeable rg
constant experiment 1 experiment 2 experiment 3 sphere β) (23)
rh
D 1.79 2.19 2.25 3
a 0.14 0.75 0.52 24
b 1.31 1.04 1.05 1 For aggregates in the size range of 500 e rh e 7000Å, Wilzius
found that β ) 1.38. This result was 2.4 times greater than
simulations predicting β ) 0.57.
macropores between clusters is likely to be much greater
Rogak and Flagan (25) computed the Stokes drag on self
than the permeablity inside the smaller clusters, resulting
similar clusters of spheres by decomposing clusters of n
in a non-uniform permeability within the porous aggregate.
monomers into smaller clusters which were replaced by
Empirical Drag Coefficients of Fractal Aggregates.
hydrodynamically equivalent spheres. Their simulations
Based on these experimental results it was possible to
predicted that β was a function of the fractal dimension
calculate drag coefficients measured in the three experi-
and the number of monomers. For D ) 1.0 (a single chain
ments assuming the drag coefficient was adequately
of particles) β ranged from 1 to 4 for N ) 102-106. For
represented by the function Cd ) aRe-b, and assuming that
larger fractal dimensions, increases in β were not as great.
all other geometrical factors were as specificed in Stokes’
For N ) 102-106, β ≈ 1.5 for D ) 1.79 and β ≈ 1.4 for D )
law. Re-deriving Stokes’ law in terms of the drag coefficient
2.1. Data from the three experiments reported here can be
and all directly measured values produced
compared with Rojak and Flagan’s results by using the
2Nvo∆F relationship from their analysis that
Cd ) (22)
AFlU2
(D D+ 2)
1/2
r g ) rf (24)
The drag coefficients, derived from the plots in Figure 7,
are summarized in Table 3. Since the slopes calculated where rf is the outer radius of the aggregate calculated
from the size-settling velocity data compared well to from the size of the smallest circle that will just fully
predicted slopes for experiments 2 and 3, there were only encompass the aggregate. The hydrodynamic radius is the
small changes in the predicted constant b for different fractal aggregate diameter calculated in Stokes’ law. Using rh )
dimensions. However, the much faster settling velocity of d/2 and eq 10, rh can be obtained from the measured

1916 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 6, 1996

 sphere. Since the measured drag coefficients of fractal
aggregates in this study were lower than those for imper-
meable spheres, reflected in lower values of a (Table 2), the
aggregate porosity constant e would be overestimated by
eq 28. If eq 28 were applied to data in our study, the
aggregate solid fraction (1 - p) would have been calculated
to be 4-8.3 times greater than it actually was in order to
account for the higher measured settling velocities than
predicted using the impermeable sphere drag coefficient.
An additional problem with eq 28 is that it leads to the
scaling relationship of U ∼ lD-1 (eq 15). Calculation of D
using this approach is only valid when D > 2. If D < 2, D2
must be included the scaling relationship (Table 2) since
D2 * 2 as it is for an impermeable sphere. If D2 is not
included, the magnitude of D will be overestimated. For
example, D calculated for experiment 1 (where D < 2) using
FIGURE 8. Dimensionless ratio (β) of the average and hydrodynamic eq 15 produces D ) 2.04 based on settling velocity data, but
radii as a function of the number of particles (N) in the aggregate: D is measured directly as D ) 1.79 from size-porosity data.
O, experiment 1, D ) 1.79; 4, experiment 2, D ) 2.19; 0, experiment Equation 28 can be modified to include D2 by using eq
3, D ) 2.25. 16 with a proportionality constant of h, to produce
parameters as
eh g∆F D-D2+1
U) d (29)
g∆FNvo 3a νFl
rh ) (25)
6πνFlU
While this equation will be correct for fractal geometry
Combining eqs 23-25, we have relationships, constants derived for eq 29 are not true
constants. The constants will likely vary (as they have in
6πνFlUrfD1/2 this study) unless the aggregates all have the same fractal
β) (26) dimension and are developed from the same primary
g∆FNvo(D + 2)1/2 particles under identical conditions. The basis for this
statement is that geometrical equations derived by Jiang
Shown in Figure 8 is our settling data expressed in terms and Logan (19) to describe aggregate morphology indicate
of β and the number of primary particles in the aggregate. that the constants e and h are functions of the fractal
The large differences in the number of beads in the two dimension. Results given in Table 3 show that a is a function
experiments with similar fractal dimensions result from of the fractal dimension as well. Thus, an empirical
the the generation of these aggregates from monomers of approach to the description of aggregate properties or
different sizes (2.6 µm diameter beads for D ) 2.19 and 0.87 settling velocity relationships should produce different
µm for D ) 2.25). While there are clearly large variations constants for eq 25 under different conditions of aggregate
in the data, calculated β’s are larger by a factor of 2.7-6.4 growth. Surveys of fractal dimensions obtained for various
than those predicted by Rogak and Flagan, indicating that types of aggregates produced under different conditions in
their simulations of hydrodynamic friction of fractal ag- engineered and natural systems vary widely (1, 2, 26-28),
gregates overestimates the drag experienced by settling supporting our speculation of a lack of a “universality” of
aggregates. aggregate fractal dimensions in these systems.
Since fractal aggregates experience less drag than Several investigators have examined aggregate settling
predicted using Stokes’ law, many previous studies have by including a shape factor, φ, into Stokes’ law, in the form
incorrectly calculated aggregate densities, porosities, pri-
mary particle densities or fractal dimensions from settling g∆F(1 - p) 2
velocity data for aggregates with fractal dimensions less U) d (30)
φ18νFl
than 2. For example, Tambo and Watanabe (16) resolved
observed settling velocities with those predicted using
The reduction of drag observed here as a result of the fractal
Stokes’ law in eq 8 by assuming that the aggregate porosity
nature of the aggregate, however, is not the same phe-
could be described according to the function
nomenon described by a shape factor. Settling velocities
of nonspherical aggregates are decreased compared to those
(1 - p) ) ed-f (27)
for spheres, not increased as observed in our studies.
A comparison of eq 27 with the fractal scaling relationship The reduced drag coefficients calculated for the fractal
in eq 14 reveals that f ) 3 - D. If we derive Stokes’ law microsphere aggregates may not be applicable to all types
explicitly in terms of the drag relationship Cd ) aRe-b, of aggregates since many aggregated particles, such as
assuming that b ) 1, and replacing the power in eq 27 by activated sludge flocs in engineered bioreactors and marine
the fractal relationship D - 3, we obtain snow in the ocean, contain a variety of polymers, filaments
and other material that can clog pores and alter the flow
3aνFlU conditions around and within settling aggregates. Despite
(1 - p) ) edD-3 ) (28)
4g∆Fd2 these possibilities, however, there is strong evidence for
wide applicability of our observations. Marine snow
Equation 28 requires the drag experienced by a settling (aggregates > 0.5 mm) exhibit fractal scaling properties for
fractal aggregate to be identical to that of an impermeable size-area and size-porosity relationships (26), and it is

VOL. 30, NO. 6, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 1917

thought that most marine snow is formed from smaller (5) Yao, K. M.; Habibian, M. T.; O’Melia, C. R. Environ. Sci. Technol.
particles that also have fractal properties (29). Alldredge 1971, 5, 1102-1112.
(6) Rajagopalan, R.; Tien, C. Am. Inst. Chem. Eng. J. 1976, 22, 523-
and Gottschalk (30) calculated that drag coefficients for 533.
marine snow were slightly higher than for impermeable (7) Hawley, N. J. Geophys. Res. 1982, 87, 9489-9498.
spheres, but they used a relatively high particle density of (8) Kajihara, M. J. Oceanogr. Soc. Jpn. 1971, 27, 158-162.
(9) Gibbs, R. J. J. Geophys. Res. 1985, 90, 3249-3251.
1.23 g cm-3 (typical of herbivore fecal pellets) in their (10) Bird, R. B.; Stewart, W. E.; Lightfoot, E. N. Transport Phenomena;
calculations rather than lower densities more typical of Wiley: New York, 1960.
bacteria, phytoplankton, or fecal pellets of planktonic (11) Matsumoto, K.; Suganuma, A. Chem. Eng. Sci. 1977, 32, 445-
447.
predators such as chaetognaths (31). As shown above, lower
(12) Masliyah, J. H.; Polikar, M. Can. J. Chem. Eng. 1980, 58, 299-
particle densities would have reduced calculated drag 302.
coefficients. Drag coefficients reported for other marine (13) Adler, P. M. J. Colloid Interface Sci. 1981, 81, 531-535.
particles, such as re-coagulated marine sediments, are much (14) Logan, B. E.; Hunt, J. R. Limnol. Oceanogr. 1987, 32, 1034-1048.
(15) Chellam, S.; Wiesner, M. R. Water Res. 1993, 27, 1493-1496.
lower than expected for impermeable spheres even when (16) Tambo, N.; Watanabe, Y. Water Res. 1979, 13, 409-419.
an upper limit for the primary particle density of 2.65 g (17) White, F. M. Viscous fluid flow; McGraw Hill: New York, 1974;
cm-3 is assumed (8, 9). p 209.
(18) Meakin, P. Adv. Colloid Interface Sci. 1988, 28, 249-331.
Particle settling velocities are widely used in calculation (19) Jiang, Q.; Logan, B. E. Environ. Sci. Technol. 1991, 25, 2031-
of carbon fluxes to the sediments in the ocean as well as 2037.
in many engineering calculations for water and wastewater (20) Stolzenback, K. Deep-Sea Res. 1993, 40, 359-369.
treatment processes. While further work is necessary to (21) Lai, R. J.; Hudson, H. E., Jr.; Singley, J. E. J. Am. Water Works
Assoc. 1975, 67, 553.
strengthen the generality of our findings to other types of (22) Happel, J. Low Reynolds Number Hydrodynamics, with special
aggregates, our study suggests that previous calculations applications to particulate media; Prentice Hall: Englewood
relating aggregate size and settling velocity based only on Cliffs, NJ, 1965; p 385.
(23) Johnson, C. P. Settling velocities of fractal aggregates. M.S. Thesis,
Stokes’ law will need to be re-examined. The primary result University of Arizona, 1995, 118 pp.
will be that the mass fluxes estimated from settling velocity (24) Wiltzius, P. Phys. Rev. Lett. 1987, 58, 710-713.
calculations will need to be changed to account for the (25) Rojak, S. N.; Flagan, R. C. J. Colloid Interface Sci. 1990, 134,
206-218.
faster settling velocities of fractal aggregates compared to
(26) Kilps, J. R.; Logan, B. E.; Alldredge, A. L. Deep-Sea Res. 1994, 41,
impermeable spheres. 1159-1169
(27) Namar, J.; Ganczarczyk, J. J. Water Res. 1994, 27, 1285-1294.
Acknowledgments (28) Logan, B. E.; Kilps, J. R. Water Res. 1995, 29, 443-453.
(29) Li, X.; Logan, B. E. Deep-Sea Res. 1995, 42, 125-138.
This research was supported by ONR Grant N00014-91-J- (30) Alldredge, A. L.; Gottschalk, C. Limnol. Oceanogr. 1988, 33, 339-
1249. 351.
(31) Dilling, L.; Alldredge, A. L. Mar. Ecol. Prog. Ser. 1993, 92, 51-58.
Literature Cited
(1) Li, D.-H.; Ganczarczyk, J. J. Environ. Sci. Technol. 1989, 23, 1385- Received for review August 14, 1995. Revised manuscript
1389. received February 5, 1996. Accepted February 7, 1996.X
(2) Logan, B. E.; Wilkinson, D. B. Limnol. Oceanogr. 1990, 35, 130-
136. ES950604G
(3) Han, M.; Lawler, D. F. J. Am. Water Works Assoc. 1992, 84, 79-
91.
(4) Valiolis, I. A.; List, E. J. Environ. Sci. Technol. 1984, 18, 242-247. X Abstract published in Advance ACS Abstracts, April 1, 1996.

1918 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 30, NO. 6, 1996