Content uploaded by Andreas Günter Weber
Author content
All content in this area was uploaded by Andreas Günter Weber on Mar 09, 2014
Content may be subject to copyright.
EG UK Theory and Practice of Computer Graphics (2006)
M. McDerby, L. Lever (Editors)
Computing Static Electricity on Human Hair
G. Sobottka A. Weber
Institut für Informatik II, Universität Bonn, Germany
Abstract
We present a framework to study electrical charge phenomena on human hair. We propose a fiber based hair model
which bases on the special theory of Cosserat Rods to overcome the well known difficulties one has to deal with
when simple particle systems are used. We show how such models can efficiently be employed in conjunction with
the fast multipole method to account for Coulomb far-field interactions. Furthermore, we extend our model such
that we can account for environmental conditions.
Categories and Subject Descriptors (according to ACM CCS): I.3.6 [Computer Graphics]: Graphics data structures
and data types
1. Introduction
Human hair modeling is an active area of research in com-
puter graphics [CK05,SVW05,WGL04,HMT01]. Many of
the hair models that have been introduced to the computer
graphics community primarily aim at speed rather than ac-
curacy. Using the naturally occurring hierarchy of single hair
fibers and hair strands the physical simulation of hair has fo-
cussed on the level of hair strands [BKCN03,BCN03]. Only
on the geometrical level the strands are then subdivided into
single fibers for visualization.
However, for more accurate hair modeling a true hierar-
chical physical simulation is necessary connecting the level
of a hair strands with the finer level of hair fibers also on the
physical level and not only the geometrical level.
A physical simulation on the level of single hair fibers is
especially important if special physical effects influencing
the overall appearance of the hair geometry have to be in-
cluded. Up to now this is a rather neglected area. Recently,
some physical models have been introduced that capture
the effects of water and styling products [WGL04]—with
a rather limited accuracy. Other very prominent physical ef-
fects like that of static electricity on human hair have not
been included in any model so far.
In an extreme form the effect is well known from an ap-
pealing experiment in physics lesson: A person puts both
hands on a Van de Graaf generator. A belt made of plastic or
rubber is turned by a motor and builds up a “static charge”
by scratching on a wire which is attached to a metal shell.
This effect is well known as triboelectricity. The negative
charge flows directly from the surface of the metal shell to
the surface of the test person. If the hair is clean and dry and
the fibers do not stick together due to lipids or moisture on
their surface they should start to lift because each fiber has
a repelling negative potential. The fibers follow the stream-
lines of the electric field and take the typical shockheaded
hair style.
Also in non-extreme cases the effects of static electricity
might visible contribute to the overall appearance of human
hair. Even normal combing or brushing causes static elec-
tricity in human hair changing the overall hair volume and
smoothness in subtle but visible ways. As there are differ-
ent effects from combing or brushing which cannot be sep-
arated empirically from each other highly accurate models
are necessary that are capable to predict the effects of static
electricity among others.
Our contributions. We introduce a fiber based hair model
which can be seen as the first step towards a high accuracy
model for hair modeling tasks in general. We employ a static
Cosserat model to model single hair fibers and demonstrate
how this approach can be combined with a multipole expan-
sion of point charges to capture the effects of static electric-
ity on human hair.
As multipole expansions are little used in computer graph-
ics we will discuss them in Sec. 2.
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
Figure 1: Potential field of a hair strand computed from dis-
crete point charges on a 128 ×128 ×128 grid (cf. results
section).
In Sect. 3we discuss an equilibrium state rod model based
on the special theory of COSSERAT rods that is suitable as a
fiber based hair model. Results of coupling the previously in-
troduced fiber based hair model with a computation of elec-
trostatic forces are given in Sect. 4.
2. General Setup for Computing Electrostatic Forces
Electrostatic problems are governed by Coulomb’s law. Let
the objects of investigation be simple point charges dis-
tributed in space. The Coulomb force acting between two
point charges located at x1and x2is given by fC=c0·
(q1·q2/|x2−x1|3·(x2−x1))†. In case of many particles the
Coulomb interaction follows directly from the principle of
superposition which is valid for electrostatic fields:
fC=c0·
n
∑
j=1,i6=j"qi·qj·xi−xj
xi−xj
3#.(1)
The main problem of computing electrostatics between a
given set of point charges is that the Coulomb mutual in-
teractions of all particles have to be taken into account even
if they are far apart from each other. Thus, thecomplexity is
O(n2).This is known as the N-body problem. Models that
account for physical influences due to sources that are far
away are normally referred to as far field models.
Problems involving many bodies are well known in the
computer graphics community as they frequently occur in
the realm of collision detection. But these problems should
not be confused with the classic N-body problem because the
bodies do not interact over a wide range. Collisions and their
†c0depends on the choice of units and is equal to 1 for the cgs
system.
subsequent responses are a local phenomenon. Thus, strate-
gies to accelerate the computation of pairwise Coulomb in-
teractions cannot rely on spatial or object subdivision tech-
niques alone like in the collision detection problem. Such
techniques only allow us to limit the sphere of action and fo-
cus on the local interactions. However, simply truncating far-
field interactions above a predetermined distance can lead to
unphysical behavior as was shown by [DYP93]. What we
need is a technique to approximate far-field interactions be-
cause intuitively the impact of two adjacent bodies on a third
body which is far away should be very similar. The missing
link to accomplish this is known in the physics community
as multipole expansion.
2.1. The Multipole Expansion
To understand the concept of the multipole expansion of
the electrostatic potential consider a point charge qat po-
sition x0∈R3. The potential in the vicinity of the particle
is given by Φ(x) = 1/˜r, where ˜r=|x−x0|is the distance
from the particle. The multipole expansion is a series ex-
pansion that expresses the potential Φ(x)in terms of its dis-
tance rfrom the origin. The adoption of spherical coordi-
nates, x= (r,θ,ϕ)and x0= (ˆr,ˆ
θ,ˆ
ϕ),leads to an expression
in terms of Legendre polynomials:
Φ=1/˜r=∞
∑
n=0ˆrn/rn+1Pn(cosγ),(2)
where γis the angle between x0and x. This, in turn, can be
expressed by a class of functions known as spherical har-
monics. These constitute a set of orthogonal solutions to the
underlying Laplace equation ∇2Φ=0 if we consider the
problem in spherical coordinates.
We now examine kcharges of strength qilocated at the
positions ˆ
x= (ˆri,ˆ
θi,ˆ
ϕi)within a sphere of radius R. The po-
tential for any point x= (r,θ,ϕ)outside the sphere is then
given by
Φ(x) = ∞
∑
n=0
n
∑
m=−n
Mm
n
rn+1·Ym
n(θ,ϕ),(3)
where
Mm
n=
k
∑
i=1qi·ˆrn
i·Y−m
n(ˆ
θi,ˆ
ϕi),(4)
is the moment of the expansion and accounts for the
charge distribution. Ym
n(θ,ϕ)are spherical harmonic poly-
nomials of degree nand defined as
Ym
n(θ,ϕ) = s(n− |m|)!
(n+|m|)!Pn|m|(cosθ)eimϕ,(5)
where Pm
n(x)are the associated Legendre polynomials.
The above equations are only valid if the particles are well
separated. Further, it is important to note that since these
approximations do not converge at close ranges one must
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
Figure 2: Left: The outer multipole expansion expresses the
potential outside a sphere of radius R due to all particles
inside the sphere. Right: Boundary conditions; Interference
of particles on opposite sides of the head.
further explicitly compute near field interactions. If the dis-
tance r>> Rof a particle is large compared to the radius
of the surrounding sphere for which the multipole expan-
sion is to be computed then we approximate the Coulomb
interaction with all the particles inside the sphere by eval-
uating a finite number pof terms in the expansion. Thus,
the forces acting at each particle consist of a fraction due
to the far field (ff) and one due to the near field interactions
(nf): fΣ=fnf +fff +fext,where ext are external forces due to
gravity or collision response etc.
The corresponding electrostatic field is given by the gra-
dient of the potential
E=−∇Φ(x) = −∂Φ/∂xi.(6)
From the electrostatic field we obtain the forces acting on
each particle as F=q·E.An example of a potential field
of a hair strand computed with our approach can be seen in
Fig. 1.
2.2. Charged Tracer Particles
The multipole expansions have been derived for point
charges and it is possible to do the same for other primitives
like cylinders or tori. But this is cumbersome and normally
not compatible with standard multipole codes. To approxi-
mate the charged surface of a thin generalized cylinder with
a given length we introduce the concept of tracer particles.
Therein, each hair fiber is equipped with a set of (negative)
point charges. We then compute a multipole expansions of
the potential to approximate the far-field interaction of each
particle. From this we can deduce the Coulomb force acting
on the particles as the gradient of the potential.
It is not clear how the particle density ρPcan be chosen
efficiently. The forces due to Coulomb interactions can be
split into a tangential part and a normal part. The tangential
part causes the tracer particles to shift along the fiber until
a equilibrium is reached. In reality this effect takes place at
the speed of light. After redistributing the tracer particles it
would be necessary to recompute the the Coulomb interac-
tions until the system of particle converges against a stable
equilibrium. However, this modus operandi can become very
expensive if the number of particles is large. In practice, we
simply choose a constant density for the tracer particles. This
density does not necessarily correspond to the node density
of the fibers so it is independent of the underlying discretiza-
tion. The tracer particles are fixed in space and do not move
as the corresponding fiber deforms. Thus, it is necessary to
map them back to the corresponding positions after each de-
formation step. For this, we introduce a bijective mapping
based on the arc length parameter. Their cartesian coordi-
nates can then be computed by linear interpolation from the
coordinates of the nodes to the left and to the right of a par-
ticle.
2.3. Multipole Codes
A number of approaches have been proposed by the physics
community to tackle the N-body problem. These approaches
employ standard space or object partitioning techniques like
octrees or bounding volume hierarchies to subdivide the
computational domain. The tree like algorithm of Barnes
and Hut [BH86] originates from the field of astrophysics and
performs at costs of O(nlogn). The Fast Multipole Method
(FMM) of Greengard and Rokhlin [GR87] is one of the most
successful as it is based on analysis of the pairwise interac-
tions of point charges and allows for a priori error bounds.
The computational costs are reduced to O(n)by introduc-
ing the concept of local expansions. The main difference
between the FMM and the algorithm of [BH86] is that the
former operates with potentials whereas the latter directly
computes forces.
2.3.1. The Fast Multipole Method
We decided to base our electrostatic computations on the
FMM. First an octree is built on the cloud of tracer parti-
cles distributed over the fibers with a given density ρP. At
the leaves of the tree the multipole expansion is computed
directly from all particles within a cell, i.e., from their posi-
tions and strengths. The computation of the potential per-
forms in two passes, one upward pass to accumulate the
multipole expansions of the cells and one downward pass to
compute the potential for each particle inside a cell due to all
particles outside the cell. Near field influences are computed
concurrently since they only depend on the nearest neigh-
bors which are located in adjacent octree cells.
The total number of operations to be performed is
189(N/s)p4+2np2+27ns,where sis the average num-
ber of particles per leaf node and pis number of terms in
the multipole expansion. In order to achieve a desired pre-
cision of εthe number of terms that must be evaluated is
p=log√31/ε.The translation and conversion of multipole
expansions can be justified by foundamental theorems of the
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
theory of spherical harmonics. We refer the interested reader
to the original work of Greengard and Rokhlin [GR87].
2.4. Particle-Surface Interactions
The scalp surface—or the object’s surface in general—
normally contributes to the overall deformation since it car-
ries its own charge. Moreover, taking into account the re-
pelling forces between the scalp and the hair fibers acts as
a form of collision handling as it prevents the fibers from
moving into the object. To allow a fast charge computation
we calculate for the head surface San implicit linear ap-
proximation, a distance field ΦSwhich is a scalar mapping
ΦS:R3→Rsuch that
ΦS(x):=min
q∈S {|x−q|},∀q∈R3.(7)
Strictly speaking, Coulomb’s law applies only to charged
particles without extension, but we still use it here for the
force computations between the particles and the scalp.
Hence, particles in the vicinity of the scalp experience a re-
pelling force which decreases with 1/r2.The force is given
as
fS(x) = |qi| · |QS|
r2
x·∇ΦS(x)
|∇ΦS(x)|,(8)
where ∇ΦS(x)is the gradient of the implicit function at
xand rxthe smallest distance of xto S.QSis the surface
charge such that qi·QS>0.The implicit approximation of
the surface is computed using the method of characteristics.
For each triangle of the underlying mesh the Voronoi regions
of the face, the edges as well as its vertices are computed.
The Voronoi region of a face forms a prism whereas that of
a convex edge forms a wedge like area. The Voronoi region
of the vertices forms a cone if the vertex is convex. We then
compute the distances of each grid point inside a Voronoi
region to the respective feature. The distances are stored on
a rectilinear grid. The signs are chosen according to the di-
rection of the normal. For a given point xthe distance to the
surface can then be calculated in terms of a tri-linear inter-
polation over the cell’s corners in which the point of interest
lies. The total costs for the Coulomb force approximation
are, thus linear in the number of particles.
In order to prevent singular forces we assume that the ini-
tial hair configuration is collision free in the sense that a) the
tracer particles are well separated (see above) and b) none
of the tracer particles lies inside the object. If for some rea-
son b) is not met the distance has a negative sign. Since the
resulting force decreases quadratically with the distance it
is possibly to small to push the particle outside the object.
In principle, these pathological cases can be detected by a
simple sign check of d.We then use a linear or even slower
decreasing force.
2.5. Boundary Conditions
Boundary conditions normally need a special handling since
the line of action between two particles may theoretically
intersect with the boundary. In our case the boundary is con-
stituted by the surface of the human head model. It is nor-
mally not possible to incorporate such boundaries into stan-
dard multipole codes while maintaining the O(n)scheme
because the multipole expansions are usually computed on
a per cell basis. In principle, those interactions could be
computed directly and subtracted from the eligible particle’s
overall charges. However, we assume that the error is very
small as it only affects particles that are close to the surface
(cf. Fig. 2). Therefore, the bending moment induced in the
vicinty of the root of a hair is small. Since the force decreases
quadratically the effect on pairs of particles with large dis-
tance is negligible.
3. Special Theory of Cosserat Rods
The modeling of long thin structures with torsional stiff-
ness has gained increased attention in the realm of computer
graphics. Catheters, cables in microsurgery simulations, and
hairs are normally not open for approaches like particle sys-
tems. It is difficult to measure local deformations like twist
with mass-spring systems or to map material properties to
them. Another important aspect is that the temporal evolu-
tion of mass-spring approximated materials with high inter-
nal viscosity leads to the well known stiff-equation prob-
lems. In the realm of hair modeling the research tends to the
development of more accurate models. So called rod theories
which describe the deformation of long thin solid structures
subject to external loads have been developed by KIRCH-
HOFF [Kir59] and later by the Brothers COSSERAT. The lo-
cal director approach allows one to accurately model local
deformations like shear, bending, and twist. Further, material
properties can directly be incorporated in terms of so called
constitutive relations. Following Antman [Ant95] the equa-
tions describing the equilibrium state of a rod can be derived
as follows.
3.1. General Setup
Given a smooth space curve r(s):s⊂R−→ R3which de-
scribes the center line of a filament in its reference state with
sbeing the arc length parameter. We equip this space curve
with a pair of orthonormal directors d1(s)and d2(s)such
that they span the cross section plane at s.This cross section
is normal to the curve at sand therefore we obtain a third
direction by d3:=d1×d2.The directors diform a right-
handed orthonormal basis for R3.Furthermore, we assume
the existence of vector-valued functions uand vsuch that
d0
k=u×dkand r0=v·dk,where (·)0denotes the derivative
w.r.t. the arc length parameter. The former is normally re-
ferred to as kinematic relation. The components of this func-
tion are given by u=ukdkand r0=vkdk.The components
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
of u:= (u1,u2,u3)and v:= (v1,v2,v3)have physical mean-
ings as they are the strain variables describing the motion of
a material cross section through space. The first two com-
ponents of uindicate the flexure about d1and d2and u3
measures the torsion, i.e. the rotation about the normal of
the cross section whereas the components of vdescribe the
shear and the expansion, respectively. Thus, the centerline C
of the fiber is determined by r(s),d1(s),and d2(s),except
for a rigid motion, i.e., rotation and translation.
Two degenerate cases can occur in our filament: 1) if we
allow the curve to expand arbitrarily it can take local zero
length, i.e., |r0|=0.This can be prevented if we ensure that
the expansion always is v3>0,thus r0·d3>0.2) If we
further allow the filament to undergo shear deformation the
cross section plane is no longer orthogonal to the centerline.
In the extreme case it is tangent to the curve. This can also
be prevented by the condition v3>0.
To derive the balance laws consider a rod element of
length ds =s+−s−,with 0 <s−<s+<l. Then the rod
element (s+,l]exerts a resultant contact force n+(s+)and
contact torque r(s+)×n+(s+) + m+(s+)about the origin
on the section [s−,s+], where m+(s+)is the resultant con-
tact couple. n+(·)and m+(·)depend only on the separat-
ing section between the rod elements [0,s−)and (s+,l].The
contact force and torque about the origin from the left rod
element exerted on [s−,s+]are denoted by −n−(s−)and
−r(s−)×n−(s−)−m−(s−).Other forces and moments
acting on [s−,s+]are conveniently expressed in the integral
form Rb
af(s)dsand Rb
a[r(s)×f(s) + l(s)]ds,where f(s)and
l(s)are the body force and couple per unit reference length.
From this we obtain the equilibrium equations as
n0+f=0(9)
m0+r0×n+g=0.(10)
Since the rod is clamped at one end (s=0) and can freely
move at the other we have the following boundary con-
ditions: x(0) = x0,d1(0) = d10,d2(0) = d20.Due to the
Coulomb forces acting at the discrete positions of the tracer
particles we also have a fixed number of boundary conditions
of the type f(si) = Fito be matched in between the end points
of a fiber. Further, the tangential part of the Coulomb forces
leads to a transversal stretching in the fiber. Using the rela-
tion that locally v= (0,0,1)we can easily solve the problem
of forces acting in transversal direction. This means that the
fiber is inextensible and unshearable an assumption that can
be justified by the fact that real hair fibers show a very high
tensile stiffness. Shear deformations are of rather limited in-
terest for us so we totally neglect them.
3.2. Solving the Cosserat Equations
The above equations together with the boundary conditions
form a system of coupled ODEs. Such a two-point boundary
value problem (BVP) can be solved with appropriate shoot-
ing techniques. They perform recurrent integrations of the
underlying ODE’s while trying to minimize the distance to
the boundary conditions with root finding techniques such
as the Newton-Raphson method. Since the equations (10)
are known to be stiff by their very nature and exhibit singu-
larities in the solution one should at least use BVP-solvers
with continuation [HM03]. Such continuation methods are
superior to standard BVP-solvers but perform at slow con-
vergence rates. Moreover, the convergence rate depends on
the quality of the initial guess. In [Pai02] Pai demonstrates
how to approximate the solution to the above equations by
decoupling the balance laws from the kinematic relations
which can be integrated explicitly by using Rodrigues’ for-
mula which is in fact equal to linearizing the rotations. This
method can be used in conjunction with standard BVP-
solvers to compute the Jacobian matrix. However, recurrent
evaluations of this method are very expensive. Moreover,
this approach operates with local forces such that each load
is transformed to a follower load, a problem which we will
discuss in detail later. In contrast, we pursue an approach
wherein the energy functional of the rod is minimized.
3.3. Parameterizing Rotations
To solve the kinematic relation d0
i=u×diit is essential
to choose an appropriate parameterization for the rotations.
We use exponential maps, an approach that was originally
introduced to the field of rod mechanics by Simo and Vu-
Quoc [SVQ88] (also used in [Pai02], [CJY02]). The expo-
nential map maps the rotational strain vector uspecifying
the rotation axis and the magnitude of a three DOF rotation
to the respective rotation. Together with the linear strain vec-
tor v(here always v= (0,0,1)) it describes a screw motion.
The relative rotation from frame F(si)to frame F(si+1)is
obtained as eˆ
wθ≈I+ˆ
wsinθ+ˆ
w2cos(1−cosθ),and the
corresponding translation as (I−eˆ
wθ)ˆ
wν+wwTνθ,where
w=u/|u|,ˆ
w=w×,ν=v/|u|,and θ=|u|·d.d=si+1−si
is the distance between two consecutive points on the curve.
3.4. Energy Minimizers
As stated before our approach is based on an energy mini-
mization scheme which provides a solution to the Cosserat
equations. Moreover, it allows us to integrate single forces
acting in-between the end point of the rod. In particular, it
seeks a rotational strain function u∗(s)that describes the
minimal energy configuration of a rod subject to external
forces. A similar model has been introduced by [WH04]. It is
based on the differential geometry of the curves and follows
the standard approach in which a space curve is equipped
with a set of orthogonal directors. The orientation at each
point is expressed in terms of global Euler angles. Since the-
ses are a function of the arc length parameter they can be
approximated by a linear combination of appropriate basis
functions (Ritz approach). Suitable coefficients can be found
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
Figure 3: Left: Sequence of deforming rods due to a terminal follower load with increasing magnitude. The direction of the
terminal force remains always perpendicular to the center line of the rod. Center: Young’s modulus, torsional modulus, weight
and diameter of human fibers depend on the humidity (isotherms). Right: Computation times (in sec) taken with the fast mul-
tipole method for varying numbers of terms in the multipole expansion (on a Pentium IV 2.4 GHZ). The dotted line shows the
corresponding mean direction deviation from the correct solution averaged over all particles.
by solving an unconstrained optimization problem wherein
the internal energy of the rod is minimized. It is normally not
an easy task to find a set of basis function which are com-
plete and can deal with the singularities caused by the use of
global Euler angles. Besides, the times needed to compute
the solution to simple problems (e.g., the straight rod under
gravity) is comparatively large. In our approach we directly
operate on the discretized data.
Our approach bases on the minimization of the internal
energy of the rod which is given as the sum of the stored
internal energy UE, the potential energy UBdue to body
forces, and the work done by all external forces acting on
the rod:
UT=UE− U B− W.(11)
Therefore, we assume for a hyperelastic rod the exis-
tence of an stored energy function Γ= (v,u,s)such that
n=∂Γ/∂vand m=∂Γ/∂u.We then obtain the elastic en-
ergy as the integral over the entire domain Ω:
UE=ZΩΓ(u)ds.(12)
The stored energy is defined by the difference between the
actual configuration and its reference configuration HC(u)−
H0(u).If we choose the stored energy function to be
Γ(u) = tr1
2C·[I(u−ˆ
u)]2,(13)
the simple constitutive laws are given as the derivatives of
this function. The quantities ˆ
umeasure the rotational strains
in the reference state. The stored energy function takes its
global minimum in the reference configuration, thus the ref-
erence configuration is stress-free. vdoes not contribute to
the energy since we have chosen the rod to be inextensi-
ble and unshearable. The Cii’s are material constants where
C11 =EI1,C22 =EI2,C33 =GI3, and Cij =0 for i6=j.E
and Gare known as Young’s modulus and the torsion modu-
lus, respectively. The quantities I1,2,3are the respective mo-
ments of inertia of the cross section. The directors d1and
d2are normally chosen such that they coincide with I1and
I2. Since these quantities depend on the humidity as well as
the temperature we have a simple model that accounts for
environmental conditions. The dependencies have been de-
termined empirically (cf. [Rob02]) and are shown in Fig. 3.
Here we focus on the isotherms, i.e., the dependency on the
humidity for a fixed temperature (25◦C). Thus, for a given
humidity we choose accordingly the stiffness parameters as
well as the weight of the fiber.
Furthermore, the potential energy due to body forces like
gravity is given by
UB=|g|ZΩl(s)r(s)g|·| ds,(14)
where lis the weight per unit length and g|·| is the unified
force field vector.
3.4.1. External Forces
In principle, external forces acting at discrete points of the
rod can be treated as shown in [Pai02]. The most promi-
nent problem with this approach is that forces are considered
w.r.t. to the local system of a node. A force that is dependent
of the rod’s configuration is called follower load. To illus-
trate the impact of follower loads let us consider a vertical
rod and a terminal force that is perpendicular to the center
line of the rod. Intuitively, we expect the rod to be deformed
in such a way that the centerline of the rod is at most paral-
lel to the force. If the force is given in local coordinates of
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
the end point then it deforms beyond the vertical shape to a
semicircle because the terminal load stays perpendicular to
the rod’s centerline (cf. Fig. 3). As we further increase the
magnitude of the force the rod assumes an s-like shape and
so on. Each turn to the opposite side adds a hunchback to
the rod. Follower loads have applications, e.g. if we want to
simulate a hosepipe ejecting water. Then the reaction force
is always tangential to the end of the hose. In case of the
Coulomb interactions we want the forces to act w.r.t. to the
global system. The work done by an external force fis given
by W=f·δx.Thus, only displacements in the direction of
the force contribute to the overall internal energy. The point
of application of a force and its displacement respectively are
obtained by integrating the kinematic relations, cf. Sec. 3.1.
In principle, fcan be any type of external force, Coulomb as
well as contact forces or cohesion. Thus, the total work done
by all external forces is
W=∑
ifi·δri.(15)
The negative signs in eq. (11) are due to the fact that ex-
ternal forces add energy to the system.
As mentioned in Sec. 2we consider the Coulomb forces
due tracer particle interactions as well as those from the in-
teraction with the scalp. Although it is possible we currently
do not consider cohesive forces that stick the fibers of a sin-
gle hair strand together.
3.4.2. Discretization
The total energy function to be minimized is given by eq.
(11). So, we seek for a minimizer to this variational problem
which describes the state where the external forces due to
gravity and Coulomb force are in balance with the material
internal reaction forces. Since each force induces local per-
turbations we discretize our domain. To solve for the equilib-
rium configuration we employ a conjugated gradient solver.
Since the rotation vector uiis given for each node the dimen-
sion of the underlying optimization problem depends on the
discretization. The Jacobian matrix ∂UT/∂uof the energy
function can efficiently be approximated by a forward finite
difference scheme. Specifically, we use an equidistant grid
with a fixed size of approximately three nodes per cm. Thus,
for a hair fiber of length 30 cm the underlying minimization
problem has approximately 300 dimensions.
4. Results
To test our approach we drove different experiments. For
this, we implemented a test environment in C++. For the
multipole expansion we used the freely available DPMTA
library which is written in C [Ran02]. Although this library
has been implemented for parallel environments we only
used the single processor mode. In order to reduce memory
costs the code was fully adapted to our simulation environ-
ment such that the computations could be performed directly
on our internal fiber data structure. All experiments and mea-
surements were carried out on a Pentium IV, 2.4 GHz ma-
chine with 2 GB of RAM.
Specifically, we considered the following test cases:
1. A curly hair tress consisting of 300 fibers (100 segments
each, length: 30 cm), uniformly distributed over an ellip-
tic cross section.
2. As a more complex example we generated a hair style
consisting of about 65,000 fibers which covered one half
of a human head model (half scalp model).
The initial configurations were generated with our
Cosserat approach. The geometric properties considered
herein correspond to values found in real human hair.
Our algorithm first distributes the tracer particles over the
fibers according to a given charge density and computes the
Coulomb interactions. Then the new configuration of the
fibers is calculated solving an unconstrained optimization
problem as described in Sec. 3.4. Specifically, we used a
tracer particle density of ρP=330 particles/m. The simu-
lation temperature was 25◦C for all computations at varying
relative humidity (65% and 100%). The charge density was
−7.7·10−4C/cm. The results can be seen in Fig. 4.
We tested two approaches for the charging: a) bringing up
the full charge in a single step and b) dividing it over nsteps
in order to simulate the built up of the electrostatic field. For
this, we assume that the amount of charge that can be ab-
sorbed by the current configuration decays logarithmically
over nsteps. It turns out that applying the full charge in a sin-
gle step does not yield the same final configurations as when
the maximum charge is applied in several subsequent steps.
Applying the full charge in a single step leads to more asym-
metric force distributions with larger forces on the outer
fibers of the strand. In the latter case the system converges
against a symmetric configuration. The cause may be that
small asymmetries in the force distributions are smoothed
out by the subsequent integration process of the Cosserat
equations. As stated in Sec. 3.4 we solve an unconstrained
optimization problem in order to find the equilibrium con-
figurations of the fibers. This optimization process directs
the global energy into a minimum and—as is known— such
minima correspond to symmetric system states.
For the strand we measured the electric field strength
given by E=F/qand averaged it over all particles. For
comparisons we consulted one of the rare publications which
contains field strength measurements on human hair, an US
Patent on ’static reducing’ hair conditioners [Pat88]. Therein
a set of pre-treated tresses were charged by combing them
with 50 strokes of a rubber comb. The sensing probe of the
fieldmeter was placed at a distance of 12 cm from the sur-
face of the tresses. The values range between 20 kV/m and
120 kV/m and correspond to ours of 13 kV/m.
Furthermore, we experimented with different numbers of
terms in the multipole expansion. As can be seen in Fig. 3
c
The Eurographics Association 2006.
G. Sobottka & A. Weber / Computing Static Electricity on Human Hair
Figure 4: The pictures show the results of our approach to electrostatics computation on human hair. 1) Curly fiber assembly
consisting of 300 individual fibers (length 30 cm, 100 segments per fiber). 2) After applying a charge (T =25◦C, RH=65%). 3)
Increasing humidity to 100%. Note that due to the weight increase and the decrease of the stiffness parameters the fibers tend
to loose their shape, an effect than can be observed in real hair. 4) Same charge but without the effect of gravity (T =25◦C,
RH=65%). 5) Half scalp model consisting of 65,000 fibers of equal length before and after applying a charge to both the scalp
surface and the hair fibers (100 segments per fiber).
the time increases almost quadratically with the number of
terms. The blue curve shows the mean direction deviation
from the exact solution obtained with the naive force com-
putation. The angles are averaged over all particles of the
strand. The deviation starts at 10−2rad for two terms and
decreases to a value of 10−9rad. The largest mean deviation
from the exact force magnitude observed was 0.3%.Finally,
we found that using more than two terms in the expansion
has no effect on the visual results.
5. Conclusion and Future Work
We presented a new paradigm which is intended as a first
step towards a high accuracy physical hair model covering
the level of single hair fibers and the level of hair strands.
This approach accurately models hair on a per fiber basis
rather than employing guides with subsequent interpolation.
As a first instance we have demonstrated how the effects of
electrostatics on human hair can efficiently be modeled by
employing multipole expansions of distributed point charges
and coupling it to an equilibrium state rod model based
on the special theory of COSSERAT rods. Currently, we are
working on numerical simulations which capture the effects
of electrostatics induced by brushing the hair. We hope that
within such a simulation framework we can also tackle the
question of the relation between the contributions of static
electricity to the observed increase of volume of hair strands
after brushing and the contributions of plastic deformation
within fibers and a higher resulting curliness within the fibers
of a hair strand.
References
[Ant95] ANTMAN S. S.: Nonlinear Problems of Elasticity, vol. 107 of Appl. Math. Sci.
Springer-Verlag, Berlin and New York, 1995.
[BCN03] BAND O Y., CHE N B.-Y., NI SHITA T.: Animating hair with loosely connected
particles. In Computer Graphics Forum (Proceedings of Eurographics 2003) (2003),
vol. 22(3).
[BH86] BARNES J., HUT P.: A hierarchical O(N log N) force-calculation algorithm.
Nature 324 (1986), 446–449.
[BKCN03] BERTAILS F., KIM T.-Y., CANI M.-P., NEUMANN U.: Adaptive wisp tree
- a multiresolution control structure for simulating dynamic clustering in hair motion.
In Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Computer
Animation (July 2003), pp. 207–213, 377.
[CJY02] CHA NG J. T., JIN J., YUY.: A practical model for hair mutual interactions.
In ACM SIGGRAPH Symposium on Computer Animation, San Antonio (2002), pp. 73–
80.
[CK05] CHO E B., KOH.-S.: A statistical wisp model and pseudophysical approaches
for interactive hairstyle generation. IEEE Transactions on Visualization and Computer
Graphics 11(2) (2005), 160–170.
[DYP93] DARDEN T. A., YORK D. M., PED ERSEN L. G.: The effect of long-range
electrostatic interactions in simulations of macromolecular crystals: A comparison of
the ewald and truncated list methods. J. Chem. Phys. 99(10):10089 (1993).
[GR87] GREENGARD L. F., ROKHLIN V.: A fast algorithm for particle simulation.
Journal of Computational Physics 73, 325 (1987), 325–348.
[HM03] HEA LEY T. J., M EHTA P. G.: Straightforward computation of spatial equilibria
of geometrically exact cosserat rods, 2003. http://tam.cornell.edu/Healey.html.
[HMT01] HADA P S., MAGNENAT-THALMANN N.: Modeling dynamic hair as a con-
tinuum. In Computer Graphics Forum (2001), vol. 20(3).
[Kir59] KIRCHHOFF G.: Über das Gleichgewicht und die Bewegung eines unendlich
dünnen elastischen stabes. J. F. Reine Angew. Math. 56 (1859), 285–313.
[Pai02] PAI D. K.: STRANDS: Interactive simulation of thin solids using Cosserat
models. In Computer Graphics Forum (2002), vol. 21(3), pp. 347–352.
[Pat88] PATEL C.: Hair conditioning composition and method. United States Patent,
Patent Number US04719104, 1988. http://210.69.13.146/search/PNUS04719104.
[Ran02] RANKIN W. T.: DPMTA - A distributed implementation of the parallel multi-
pole tree algorithm - Version 3.1. Department of Electrical Engineering, Duke Univer-
sity, 2002. Available at http://www.ee.duke.edu/~wrankin/Dpmta/.
[Rob02] ROBBINS C.: Chemical and Physical Behavior of Human Hair, 4 ed. Springer-
Verlag, New York, 2002.
[SVQ88] SIMO J. C ., VU-QUOC L .: On the dynamics in space of rods undergoing large
motions – a geometrically exact approach. Computer Methods in Applied Mechanics
and Engineering 66 (1988), 125–161.
[SVW05] SOBOTTKA G., VARNIK E., WEBE R A.: Collision detection in densely
packed fiber assemblies with application to hair modeling. In CISST’05—The 2005
International Conference on Imaging Science, Systems, and Technology: Computer
Graphics (Las Vegas, USA, June 2005), Arabnia H. R., (Ed.), CSREA Press, pp. 244–
250.
[WGL04] WARD K., GALO PPO N., LIN M. C .: Modeling hair influenced by water and
styling products. In Proceedings of the Computer Animation and Social Agents 2004
(2004).
[WH04] WAKAMATSU H., HIRAI S.: Static modeling of linear object deformation
based on differential geometry. The International Journal of Robotics Research 23, 3
(2004), 293–311.
c
The Eurographics Association 2006.
