Modeling Electrospinning of Nanofibers

T.A. Kowalewski, S. Barral and T. Kowalczyk

Abstract A fast discrete model for the simulations of thin charged jets produced
during the electrospinning process is derived, based on an efficient implementation
of the boundary element method for the computation of electrostatic interactions
of the jet with itself and with the electrodes. Short-range electrostatic forces are
evaluated with slender-body analytical approximations, whereas a hierarchical force
evaluation algorithm is used for long-range interactions. Qualitative comparisons
with experiments is discussed.

1 Motivation

Electrospinning is a simple and relatively inexpensive mean of producing nanofibers

by solidification of a polymer solution stretched by an electric field. Such fibers find
applications in a variety of areas, including wound dressing [4], drug or gene de-
livery vehicles [19], biosensors [16], fuel cell membranes and electronics [15]. Re-
cently electrospinning has been revitalized and successfully applied to the produc-
tion of nanofibrous scaffolds for tissue-engineering processes [2], which constitute
one of its most promising application.
A conventional electrospinning setup consists of a spinneret with a metallic nee-
dle, a syringe pump, a high-voltage power supply and a grounded collector. A poly-
mer solution is loaded into the syringe and driven through the needle at a steady
and controllable feed rate by the pump, forming a droplet at the tip of the needle. A
high voltage (typically up to 30 kV) is applied between the tip and a grounded col-
lector. When the electric field strength overcomes the surface tension of the droplet
an electrified liquid jet is formed. The jet is then elongated and whipped continu-
ously by electrostatic repulsion (bending instability), describing a chaotic spiraling
motion on its way to the collecting electrode. Although the process may appear

T.A. Kowalewski, S. Barral and T. Kowalczyk

IPPT-PAN, Swietokrzyska 21, Warsaw, Poland; e-mail: {tkowale,sbarral,tkowalcz}@ippt.gov.pl

1
2 T.A. Kowalewski, S. Barral and T. Kowalczyk

simple the achievement of stable operation is not an easy task. Properties that are
known to significantly affect the electrospinning process are the polymer molecular
weight, the molecular-weight distribution, the architecture (branched, linear, etc.) of
the polymer, as well as the rheological and electrical properties of the solution (vis-
cosity, conductivity, surface tension, etc.). In addition, the operating conditions such
as electric potential, flow rate, distance from the needle tip to the collection plate,
ambient parameters (temperature, humidity), and geometry of the collecting target
play a crucial role in controlling the electrospinning characteristics [1, 17, 23, 24].
Because each material demands a different optimization procedure, the develop-
ment of theoretical and numerical models of electrospinning appears thus highly
desirable.
The physical and mathematical description of the electrospinning process re-
mains, however, a distant target. Despite several parametric studies performed in
various experimental configurations [1, 11, 23] it appears difficult to formulate con-
sistent scaling laws for electrospinning. The influence of the electric field and of
the solution conductivity on the fibers quality was exhaustively investigated by
Arayanarakul et al. [1], who concluded that higher polymer solution conductivity
improves the spinning process but has only a marginal impact on the fiber size.
Broadly speaking, higher applied electric potential, higher electric conductivity and
viscosity of the polymer solution appear to improve fiber uniformity. Although the
great role played by the electric field and the solution conductivity is conform to the
intuition, experiments performed with similar configurations but different polymer
solutions exhibit at times opposite relationships [1,17]. The validation of theoretical
models with experimental data is thus far from trivial [22], and it has proved so far
impossible to relate actual fiber morphology to the predicted characteristics [8].
The early stage of electrospun jets has been investigated in several one-
dimensional theoretical and numerical works [6, 9, 10, 13, 21], which have empha-
sized the importance of viscoelastic rheology. By contrast, very few studies have
considered the unstable part of the jet, in part because of the difficulties inherent
to its three-dimensional and unsteady character. Hohman et al. [13] have derived a
linear stability theory aimed at predicting the onset of a bending mode in a straight
jet, but its relevance to the actual phenomenon remains difficult to assess. A qualita-
tive description of this instability has been given by Yarin et al. [24], using a simple
discrete model consisting of point charges connected by dumbbell elements. The
latter model appears to provide a reasonable explanation for the spiraling motion of
the jet, but suffers from mathematical inconsistencies incurred by the discretization
of the fiber into point-charges. Although this problem was remedied by account-
ing for the actual electrostatic form factors between two interacting sections of a
charged fiber [17], electrostatic interactions have usually been accounted for via
strong approximations [9, 10, 13]. One serious concern relates to the evaluation of
short-range interactions, which in the case of standard discrete integration methods
require very dense grids due to the large contribution of short-range electrostatic
interactions within distances of the order of the fiber radius [14, 17]. The fiber ra-
dius being about 103 –105 times smaller than the macroscopic scales of interest, it
appears most desirable to devise a discrete model that exploits the slenderness of
Modeling Electrospinning of Nanofibers 3

the fiber to evaluate short-range interactions in an efficient manner. Likewise, the

computation of long-range electrostatic interactions can easily become intractable
due to the O(n2 ) operation count for a pairwise evaluation of interactions between n
elements. A third issue with current numerical models is the assumption of a static
external electric field, whereas in reality the external field is modulated by the net
space charge of the fiber so as to keep constant the potential over the electrodes.
This latter issue can be effectively addressed by use of a boundary element method
(BEM). In essence, the boundary element method is a statement of the electrostatic
problem (Poisson equation) in terms of boundary integrals; as such, it involves only
the discretization of boundary surfaces, which in our case would be the electrode
surfaces and the outer shell of the fiber. Although BEM-based models have been in-
vestigated in the context of electrospraying [5,14,18,25], no attempt has been made
to cure the aforementioned inefficient evaluation of short-range and long-range in-
teractions. An efficient handling of long-range interactions would, however, greatly
benefit to both electrospraying and electrospinning simulations. For this, a vari-
ety of mesh-based and particle-based methods exist that can theoretically achieve
O(n log n) operation count, or even O(n) for the fast multipole method (FMM).
Particle-based methods are clearly at an advantage in the problem at hand, where
the computation domain is three-dimensional but is only sparsely populated by a
one-dimensional object (the fiber). Most fast particle-based algorithms are based
on either treecode hierarchical algorithms [3] or on the FMM [12]. In either cases,
distant particles are clustered and approximated as a truncated multipole expansion
(of fixed order in the case of the treecode method, and of adaptive order in the
FMM). The treecode algorithm considers particle-cluster interactions and achieves
O(n log n) complexity, while FMM considers cluster-cluster interactions and can
compute forces with O(n) complexity. At low n, however, and when the required
accuracy on the computed forces is low, the treecode is usually found to outperform
FMM due to a much smaller prefactor. Furthermore, the theoretical advantage of
the O(n) force evaluation of the FMM is somewhat mitigated by the fact that for
non-uniform spatial distribution of particles – such as a fiber in 3D space – the
charge clusterization step still requires O (n log n) operations. Last and foremost,
the treecode algorithm is simpler to implement and parallelize. It is worth noting
that treecode accelerated BEMs have been recently investigated in other contexts,
such as plasma physics [7].

2 Experiment

Electrospinning was performed inside a custom-made polycarbonate chamber of ap-

proximately 1 cubic meter volume. The spinneret consists of a 2 mm long grounded
syringe needle of internal diameter 0.35 mm, mounted vertically on an electrically
insulated stand placed 15 cm above the collecting electrode. The spinneret needle
was positively biased with a high voltage power supply. A flat cooper grid (310 mm
× 240 mm) with a small cage (75 mm × 80 mm × 50 mm) was used as ground
4 T.A. Kowalewski, S. Barral and T. Kowalczyk

electrode, as described in previous reports [16,17]. The voltage applied between the
spinneret and the ground electrode was varied from 1 kV up to 30 kV. The capillary
needle was connected through PTFE tubing to a plastic syringe filled with the spin-
ning solution. A constant volume flow rate was maintained using the syringe pump.
In our study, a 3% solution of poly(ethylene oxide) (PEO) of molecular weight
400 kDa in 40% ethyl alcohol/water mixture was used to analyze electrospinning
process. The effect of solution electrical conductivity was investigated with three
inorganic salts (NaCl, LiCl, NH4 Cl) individually dissolved in the PEO solution to
attain molar salt concentration in the range 0.001–0.02 M, which corresponds to a
15-fold variation of the solution conductivity.
The electrospinning process was observed using high speed camera (hs1200.pco)
and analyzed by evaluating the geometry of the fiber coils created by the electrified
polymer jet. The spiral envelope angle, θ , and the length of the straight jet segment,
L, were obtained from the high-speed camera images and their dependence on the
applied voltage and solution composition was evaluated and correlated with micro-
scopic images of the electrospun nanofibers.
The pure PEO solution used in the experiments exhibits a bending instability
threshold at about 3kV. Farther increase of the voltage elongates the straight part of
the jet and leads to the development of relatively stable, large amplitude looping.
The straight part increases from about 0.5 cm to over 2.5 cm. The looping cone
initially increased with the voltage but above 5kV the spiral envelope angle system-
atically decreased. It is in agreement with our previous findings [17]. This general
trend usually did not change after increasing the electrical conductivity of the solu-
tion (compare Table 1). A similar behavior of the jet could be observed for nearly
all investigated PEO-inorganic salt solutions, i.e. the straight portion of the jet in-
creased and the looping cone decreased with the voltage. However, increased salt
concentration (and hence solution conductivity) only slightly decreased the cone
angle θ for two salts NaCl and NH4 Cl, whereas the opposite behavior was observed
for LiCl. The jet straight part L increased with the salt concentration only for NaCl,
whereas it decreased for the two other salts. Relatively small changes of the jet spi-
ralling geometry may suggest simmilar elongation rate of the fiber. In fact, images
of collected fibers analysed under optical and TEM microscopes could not give any
clear evidence that solution conductivity changes fiber size or morphology. Disper-
sion of the collected data underline complexity of the problem. Increasing solution
conductivity by salt additives likely modifies other polymer properties such as sur-
face tension, viscosity, and rheology. However, according to the data collected by
Arayanarakul et al. [1] and in regard to the range of salt concentration used, only
a slight decrease of the surface tension (from 31 to about 28 mN/m) and viscosity
(from 206 to 190 mPa/s) can be expected. Hence, it is difficult to draw definite con-
clusions on the effects of solution conductivity, i.e. charge density carried by the
fiber.
Modeling Electrospinning of Nanofibers 5

Table 1 Variation of jet straight part L and cone angle θ for different salt concentrations c (and
corresponding conductivity K) and applied voltage Ug ; data for 3% PEO water/alcohol solution at
25◦ C.
c [M]
# salt Ug [kV] θ [grad] L [mm]
K [µS/cm]
3 85 7.1
5 112 23.1
0.0
1 – 10 71.3 25.2
17.1
15 63.3 17.4
20 58.7 11.8
10 86.7 26.5
0.01
2 NaCl 15 62.7 38.0
399.5
20 52.7 39.7
10 61.7 24.6
0.02
3 NaCl 15 59.0 34.6
741.3
20 50.0 42.5
10 70.3 34.2
0.01
4 LiCl 15 48.3 40.2
280.6
20 39.3 46.6
10 71.7 22.8
0.02
5 LiCl 15 63.3 10.5
551.9
20 69.3 18.0
10 75.7 32.1
0.01
6 NH4 Cl 15 64.3 32.7
457.8
20 59.0 33.8
10 84.0 24.0
0.02
7 NH4 Cl 15 71.7 30.5
906.1
20 60.0 31.2

3 Model

3.1 Governing Equations

The model is, essentially, a time-dependent three-dimensional generalization of

known slender models [6, 9, 10, 13], with the following differences:
(i) the electric field induced by the generator and by the charges on the fiber is
explicitly resolved, instead of being approximated from local parameters [6, 9,
10].
(ii) electrical conductivity is neglected. Indeed, the convection of surface charges
is believed to strongly overcomes bulk conduction at locations distant from
the Taylor cone by a few fiber radii [6]; since we are mostly interested in the
description of the bending instability, this assumption appears reasonable.
6 T.A. Kowalewski, S. Barral and T. Kowalczyk

Fig. 1 Fiber discretized into finite length elements Qk+1/2 with collocation nodes Nk localized at
the element interfaces.

(iii) for the sake of simplicity, polarization effects are neglected.

(iv) the polymer is considered a viscoelastic Maxwell fluid, as in [20].
Evaporation has been systematically neglected in the literature, except in [24] where
an ad-hoc evaporation model was used. Following Fridrikh et al. [11], we shall also
neglect evaporation even in the whipping part of the jet, by arguing that it occurs
mostly after the fiber has been maximally stretched. The mass conservation equation
accordingly reads
∂a 2 ∂(va 2 )
+ = 0, (1)
∂t ∂ξ
where a is the fiber radius, v is the velocity of the fiber and ξ is the arc length along
the fiber.
The force balance per unit of fiber length reads in turn

∂(πa 2 τ t̂ + πγ a t̂)
ρπa 2 ẍ = + λE, (2)
∂ξ
where ρ is the mass density, ẍ is the acceleration vector, γ is the surface tension
coefficient, t̂ is the local unit vector tangent to the fiber, λ is the linear charge den-
sity, ¯ is the dielectric constant of the medium (air), E is the electric field and τ
is the viscoelastic stress. The stress is given by a Maxwell viscoelastic constitutive
equation:

τ
τ̇ = G ε − , (3)
µ
where is G the elastic modulus, µ the fluid viscosity and ε the Lagrangian axial
strain
∂ ẋ
ε≡ · t̂. (4)
∂ξ

3.2 Discretization and Electrostatic Solver

A Lagrangian discrete model is used. The fiber is first decomposed into discrete
charged elements Qi+1/2 , the length of which is typically much greater than the
Modeling Electrospinning of Nanofibers 7

fiber radius, but smaller than other characteristic lengths of interest (such as the cur-
vature radius). The equation of motion (2) is then resolved at the interfaces of each
element (collocation nodes Ni , see Figure 1), where all forces are priorly evaluated.
After displacements have been calculated, the radius in the central sections of el-
ements Qi+1/2 , is straightforwardly obtained since the volume of each element is
conserved. Likewise, the linear charge density in the center section is computed by
assuming that the total charge of each element is conserved. The scheme is thus
intrinsically mass and charge conserving (but is not momentum conserving).
The forces at each collocation node are computed from the following discrete
form of Eq. 2:

1 π(a 2 τ )i+1/2 − π(a 2τ )i−1/2 + πγ (ai+1/2 − ai−1/2 )

ẍi = t̂i
2 (|xi − xi−1 | + |xi+1 − xi |)
ρπai2 1


1 γ
+ τi + κi n̂i + λi Ei . (5)
ρ ai

The values ai , τi , λi required at the collocation nodes are linearly interpolated from
those computed in the central section of the neighboring elements. The local tangent
vector t̂i and curvature vector κ n̂i are computed from the approximate osculating
circle defined by (Ni−1 , Ni , Ni+1 ).
Time integration is realized with a classical leapfrog scheme, which presents
the advantage over other second order schemes by requiring only one evaluation of
force term per time step:
n+1/2 n−1/2
ẋi = ẋi + ẍni t, (6)
n+1/2
xn+1
i = xni + ẋi t, (7)

where upper indices refer to time. Eq. (5) shows that the accelerations ẍni can be ex-
plicitly obtained from the positions xni at the same instant, except for the contribution
of the viscoelastic stress as the latter integrates a memory effect. A discretization of
the stress equation (3) which preserves second order time accuracy of the scheme
can nonetheless be found using
n+1
σi+1/2 − σi+1/2
n
n+1/2 1 n n+1
= G εi+1/2 − (σ + σi+1/2 ), (8)
t 2µ i+1/2
n+1 n+1 n+1/2 n+1/2
n+1/2 1 (xni+1 − xni + xi+1 − xi )(ẋi+1 − ẋi )
εi+1/2 = , (9)
2 (xni+1 − xni + xn+1 n+1 2
i+1 − xi )
n+1
since Eq. (8) is easily recast into a time-explicit expression for σi+1/2 .
Dynamic refinement is used in simulations to maintain the size of elements below
a prescribed characteristic length max ; whenever an element is elongated beyond
max , it is split into two elements, each containing half the charge and mass of the
initial element.
8 T.A. Kowalewski, S. Barral and T. Kowalczyk

Fig. 2 Idealized electrostatic configuration: the potential is prescribed between the tip of the needle
and a grounded infinite plane. The needle electrode is modeled as a point electrode, the charge of
which is computed such as to always maintain the prescribed potential at the needle tip. Image
charges are used to implement the potential condition φ = 0 on the infinite plane.

The needle and the grounded collector are idealized by a point-charge/plate ca-
pacitor configuration, whereas the point-charge is meant to qualitatively reproduce
the field lines concentration that exists close to the needle. The potential is pre-
scribed between the location at which the fluid is introduced (the tip of the needle)
and the infinite plane where the fiber is collected. A practical way to resolve the
electrostatic field in such a configuration is the method of images (see Figure 2),
whereas fictive mirror charges are placed symmetrically to the ground plane. The
charge of the capacitor is computed at any instant to satisfy the prescribed potential
φ = Ug at the inlet.
The electrostatic field induced by the fiber can be computed by considering that
the charges are distributed on the centerline of the fiber, as demonstrated by Homann
et al. [13]. This alleviates the need for complex form factors, as used previously by
Modeling Electrospinning of Nanofibers 9

the authors and others [5, 14, 17]. Even so, the computation of electrostatic interac-
tions between the discrete charge elements Qi remains a complex problem. First, if
a typical BEM implementation with pairwise evaluation of electrostatic interactions
is used, the problem scales as O(n2 ) where n is the number of discrete elements. A
more efficient, albeit more complex method is the hierarchical force calculation al-
gorithm (treecode) [3], which complexity scales as O(n log n). Charge elements are
in that case recursively clustered and the monopole coefficients (charge and center
of charge) of the clusters are computed. The field at a location Ni is then computed
by considering only the largest clusters which are sufficiently well separated [12],
that is, for which the distance di,C between the collocation node Ni and the charge
center of the cluster satisfies
RC
< α, (10)
di,C
where α < 1 determines the accuracy of the force evaluation and RC is the radius
of a sphere containing all the charges of the cluster. Value α = 0.8 was used in our
computations.
A typical implementation of the treecode or FMM involves the recursive decom-
position of the space domain into cubic cells, where each parent cell contains up to
8 children cubic cells, until each cell contains a single charge. Hence, the root of
the oct-tree contains all charges of the domain and its leaves contain the charges. In
our case, however, it is possible to create a hierarchized tree at a much lower costs,
since the fiber approximately organizes the charges by nearest neighbors. At each
time step, a binary tree is thus constituted by recursively grouping neighbors two
by two, as illustrated by Figure 3, calculating at each level the smallest enclosing
spheres that contain the cluster pairs. The root of the tree is a sphere that contains
all charges, and its leaves are the elements Qi+1/2 . Unlike the classical treecode
clusterization algorithm, this method creates a fully balanced tree and consequently
easily lends itself to parallelization.
To evaluate the field at a given collocation node, condition (10) is first tested on
the root cluster, and, whenever it fails, on the two subclusters recursively until clus-
ters which are sufficiently separated are found. If a leaf element is reached which
does not satisfy condition (10), the interaction is considered a close interaction and
near-field expressions must be used instead of the monopole approximation. Oth-
erwise, far field expression are used based on the monopole approximation of the
cluster. It must be underlined, however, that since charges are assumed to lay on the
centerline but induce a field on the outer shell of the fiber, one must account for a
small correction to the classical potential and Coulomb’s law:
1 qC
φC→i =  , (11)
4π ¯ d 2 + a 2
i,C i
1 qC
EC→i = u
2 + a 2 C→i
, (12)
4π ¯ di,C i
10 T.A. Kowalewski, S. Barral and T. Kowalczyk

Fig. 3 Illustration of the clustering method used in our treecode implementation: neighbor charges
are recursively grouped two-by-two and the smallest enclosing spheres are calculated at each clus-
tering level, until all charges are contained in a binary tree which root is a sphere that contains all
charges.

which is to be contrasted with the laws in 1/d and 1/d 2 for monopole-monopole
interaction. In the above, q stands for the total charge of the cluster and u for the
unit vector pointing to the collocation node. Obviously, the above formulae are exact
only for a monopole that lies along the axis of the ring where the field is evaluated.
However, the slenderness of the fiber implies that nearby charges are located close
to the axis; therefore, this condition is violated only when the cluster is far (d  a),
in which case the error made is negligible since the formula reduces to Coulomb’s
law (which is appropriate when d  a).
As mentioned earlier, condition (10) may not be fulfilled by nearby elements.
This happens in particular for elements Qi−1/2 and Qi+1/2 , which are contiguous
to node Ni . In such case, the following near-field analytical expressions are used:



λC ξi − ξC0 ξi − ξC1
C→i = asinh − asinh , (13)
4π ¯ ai ai

⎡ ⎤
ξC1 − ξC0 λC ξC1 − ξC0 λC
⎢ 1+ 1− ⎥
λC ⎢ ⎢ 2 λC 2 λC ⎥⎥
(E · t̂)C→i = −
4πai ¯ ⎢
⎣
2
2 ⎥
ξi − ξC1 ξi − ξC0 ⎦
1+ 1+
ai ai
dλC 


dξ ξi − ξC1 ξi − ξC0
+ asinh − asinh , (14)
4π ¯ ai ai
Modeling Electrospinning of Nanofibers 11
⎡ ⎤
ξi − ξC1 ξi − ξC0
⎢ ⎥
κλC ⎢ ai ai ⎥
(E · n̂)C→i = ⎢ − ⎥
4π ¯ ⎢
2
2 ⎥
⎣ ξi − ξC1 ξi − ξC0 ⎦
1+ 1+
ai ai



κλC ξi − ξC0 ξi − ξC1
+ asinh − asinh , (15)
4π ¯ ai ai

which are leading order approximation for the potential and field generated by a
slender, curved line charge on a neighboring ring. Here, λC is the linear charge
density of the line charge, ξi the curvilinear location of the location where the field
and potential are evaluated, ξC0 and ξC1 the locations of the end points of the line
charge and κ the curvature of the fiber.
Boundary conditions implemented at the inlet (tip of the needle) prescribe the
volume flow rate Qv , the surface charge density σ0 and the initial fiber radius a0 .
The initial stress is set to zero. It must be noted, however, that boundary conditions
are notoriously difficult to implement in a consistent way in electrospinning models.
This is even more true in the present case where electrical conductivity is neglected
and where the charge density is not computed self-consistently. We observe a clear
influence of both the boundary conditions and the discretization parameters at the
inlet where strong gradients take place, which clearly challenge the assumption of
slenderness. Additional investigations will be required in the future to derive bound-
ary conditions that avoid such singularities.
A small random perturbation to the position of each element introduced at the
inlet is imposed, so as to initiate the bending instability. The magnitude of this per-
turbation has no notable influence on the simulation results, provided that it is small
enough.

3.3 Simulation Results

Behavior of the code was tested using parameters typical for the experiment. Several
test runs performed confirmed general ability of the code to replicate our previous
findings [17], i.e. increased electrical potential, solution viscosity and elastic mod-
ulus decreased jet sweeping amplitude, effectively seen as decreasing of the spiral
cone. Here, we show an example of the simulation results obtained by varying fiber
charge density and the voltage. Although it is not possible to directly integrate the
liquid electrical conductivity into the model at this point, we shall assume that the
conductivity mainly affects the electrical current and thus the charge density carried
by the fiber. The simulations shown were performed for the following process pa-
rameters: surface tension γ = 0.02 N/m, elastic modulus G = 10 kPa, viscosity
µ = 10 Pa·s, mass density ρ = 1000 kg/m3, volume flow rate Qv = 1 mm3 /s,
initial fiber radius a0 = 150 µm, distance between needle tip and collector 20 cm,
and distance between point charge and needle tip 1 cm. Typical jet paths are plotted
12 T.A. Kowalewski, S. Barral and T. Kowalczyk

Fig. 4 Simulated jet paths at Ug = 10 kV (upper row) and 15 kV (bottom row), for two charge
densities 10 C/m3 (left) and 20 C/m3 (right).

on Figure 4. As we may find the model is unable to predict initial straight part of the
jet. Hence we may only compare geometry of the spiral cone.
We may conclude that higher charge density evidently increases amplitude of
the jet sweeps. It is only partly consistent with the experimental findings (compare
Table 1), where the cone angle first increases for lower salt concentrations but again
decreases for higher values. Obviously, secondary effects due to the salt-polymer
interactions may play a role for higher concentrations.
Modeling Electrospinning of Nanofibers 13

4 Conclusions

In this paper we have investigated dynamics of electrospun fibers within the slender-
body approximation proposing numerically consistent model based on hierarchical
charge clustering. A versatile boundary value method is implemented to enforce
fixed-potential boundary conditions, allowing realistic electrode configurations to
be investigated. The model is tested against experiments performed for different
polymer-salt solutions. Due to the several experimental and theoretical limitations
the model outcome predicts only qualitatively observed experimentally dependence
of the jet geometry on the solution conductivity. Father investigations are necessary
to elucidate sources of inconsistencies between the experimental findings and the
modeling.

Acknowledgements This work was supported by the Polish Ministry of Science Grant No.
N508 031 31/1740. The authors thank Diana Lamparska for her invaluable help in experimental
work.

References

1. Arayanarakul, K., Choktaweesap, N., Aht-ong, D., Meechaisue, C., Supaphol, P.: Effects of
poly(ethylene glycol), inorganic salt, sodium dodecyl sulfate, and solvent system on electro-
spinning of poly(ethylene oxide). Macromol. Mater. Eng. 291, 581 (2006).
2. Arumuganathar, S., Jayashinghe, S.N.: Living scaffolds (specialized and unspecialized) for
regenerative and therapeutic medicine. Biomacromolecules 9, 759 (2008).
3. Barnes, J., Hut, P.: A hierarchical O(NlogN) force-calculation algorithm. Nature 324, 446
(1986).
4. Bhattarai, S.R., Bhattarai, N., Yi, H.K., Hwang, P.H., Kim, H.Y.: Novel biodegradable elec-
trospun membrane: Scaffold for tissue engineering. Biomaterials 25, 2595 (2004).
5. Carretero-Benignos, J.A.: Numerical simulation of a single emitter colloid thruster in the pure
droplet cone-jet mode. Ph.D. thesis, Massachusetts Institute of Technology (2005).
6. Carroll, C.P., Joo, Y.L.: Electrospinning of viscoelastic Boger fluids: Modeling and experi-
ments. Phys. Fluids 18, 053102 (2006).
7. Christlieb, A.J., Krasny, R., Verboncoeur, J.P., Emhoff, J.W., Boyd, I.D.: Grid-free plasma
simulation techniques. IEEE. Trans. Plasma Sci. 34, 149 (2006).
8. Dayal, P., Kyu, T.: Dynamics and morphology development in electrospun fibers driven by
concentration sweeps. Phys. Fluids 19, 1011061 (2007).
9. Feng, J.J.: The stretching of an electrified non-Newtonian jet: A model for electrospinning.
Phys. Fluids 14, 3912 (2002).
10. Feng, J.J.: Stretching of a straight electrically charged viscoelastic jet. J. Non-Newtonian Fluid
Mech. 116, 55 (2003).
11. Fridrikh, S.V., Yu, J.H., Brenner, M.P., Rutledge, G.C.: Controlling the fiber diameter during
electrospinning. Phys. Rev. Lett. 90, 144502 (2003).
12. Greengard, L., Rokhlin, V.: A fast algorithm for particle simulations. J. Comput. Phys. 73, 325
(1987).
13. Hohman, M.M., Shin, M., Rutledge, G., Brenner, M.P.: Electrospinning and electrically forced
jets. I. Stability theory. Phys. Fluids 13, 2201 (2001).
14. Khayms, V.: Advanced propulsion for microsatellites. Ph.D. thesis, Massachusetts Institute of
Technology (2000).
14 T.A. Kowalewski, S. Barral and T. Kowalczyk

15. Ko, F.K., El-Aufy, A., Lam, H.: Wearable Electronics and Photonics, chap. Electrostatically
Generated Nanofibres for Wearable Electronics, pp. 13–40. Woodhead Publishing Ltd., Cam-
bridge, UK (2005).
16. Kowalczyk, T., Nowicka, A., Elbaum, D., Kowalewski, T.A.: Electrospinning of bovine serum
albumin. Optimization and use for the production of biosensors. Biomacromolecules (2008)
(to appear).
17. Kowalewski, T.A., Blonski, S., Barral, S.: Experiments and modelling of electrospinning
process. Bull. Polish Acad. Sci. 53, 385 (2005).
18. Lopez-Herrera, J.M., Ganan-Calvo, A.M., Perez-Saborid, M.: One-dimensional simulation of
the breakup of capillary jets of conducting liquids. Application to E.H.D spraying. J. Aerosol
Sci. 30, 895 (1999).
19. Luu, Y.K., Kim, K., Hsiao, B., Chu, B., Hadjiargyrou, M.: Development of a nanostructured
DNA delivery scaffold via electrospinning of PLGA and PLAPEG block copolymers. J. Con-
trol. Release 89, 341 (2003).
20. Reneker, D.H., Yarin, A.L., Fong, H., Koombhongse, S.: Bending instability of electrically
charged liquid jets of polymer solutions in electrospinning. J. Appl. Phys. 87, 4531 (2000).
21. Spivak, A.F., Dzenis, Y.A.: Asymptotic decay of radius of a weakly conductive viscous jet in
an external electric field. Appl. Phys. Lett. 73, 3067 (1998).
22. Thompson, C.J., Chase, G.G., Yarin, A.L., Reneker, D.H.: Effects of parameters on nanofiber
diameter determined from electrospinning model. Polymer 48, 6913 (2007).
23. Yang, Y., Jia, Z., Li, Q., Guan, Z.: Experimental investigation of the governing parameters in
the electrospinning of polyethylene oxide solution. IEEE Trans. Dielectr. Electr. Insul. 13, 580
(2006).
24. Yarin, A.L., Koombhongse, S., Reneker, D.H.: Bending instability in electrospinning of
nanofibers. J. Appl. Phys. 89, 3018 (2001).
25. Yoon, S.S., Heister, S.D., Epperson, J.T., Sojka, P.E.: Modeling multi-jet mode electrostatic
atomization using boundary element methods. J. Electrostat. 50, 91 (2001).