PHYSICAL REVIEW E 91, 043011 (2015)

Magnetization directions and geometries of helical microswimmers

for linear velocity-frequency response
Henry C. Fu,* Mehdi Jabbarzadeh, and Farshad Meshkati
Department of Mechanical Engineering, University of Nevada at Reno, Reno, Nevada 89557, USA
(Received 10 September 2014; published 17 April 2015)

Recently, there has been much progress in creating microswimmers or microrobots capable of controlled
propulsion in fluidic environments. These microswimmers have numerous possible applications in biomedicine,
microfabrication, and sensing. One type of effective microrobot consists of rigid magnetic helical microswimmers
that are propelled when rotated at a range of frequencies by an external rotating magnetic field. Here we focus on
investigating which magnetic dipoles and helical geometries optimally lead to linear velocity-frequency response,
which may be desirable for the precise control and positioning of microswimmers. We identify a class of optimal
magnetic field moments. We connect our results to the wobbling behavior previously observed and studied in
helical microswimmers. In contrast to previous studies, we find that when the full helical geometry is taken
into account, wobble-free motion is not possible for magnetic fields rotating in a plane. Our results compare
well quantitatively to previously reported experiments, validating the theoretical analysis method. Finally, in the
context of our optimal moments, we identify helical geometries for minimization of wobbling and maximization
of swimming velocities.

DOI: 10.1103/PhysRevE.91.043011 PACS number(s): 47.61.−k, 47.63.Gd, 47.63.M−, 87.85.gf

I. INTRODUCTION ω̂ controls the swimming direction (along ω̂), while the

frequency ω is most often used to control the speed. Typically,
Micro- and nanorobots capable of controlled propulsion
the magnetic field is rotated in a plane perpendicular to its
have been pursued for a variety of microfabrication and
rotation axis, and we focus on that case here.
biomedical applications. Biomedical applications include drug
One desirable feature for control is a linear relationship
delivery [1,2], tissue manipulation [3–5], and in vivo diagnos-
between velocity and frequency [37]. This allows simple
tics and sensing [6–9]. A number of types of microrobots em-
proportional control of the swimming velocity as frequency is
ploying different propulsion techniques have been developed,
varied and is especially useful for precise control if the linear
including chemically powered microrobots dependent on
relationship extends all the way to zero velocity. One way to
external fuels to create phoretic flows [10–18], externally con-
achieve a linear velocity-frequency relationship is to maintain
trolled biotic systems [19], dielectrophoretically manipulated
a constant axis of rotation as frequency changes [35,37], but
robots [20], and magnetically actuated robots, including those
for generic geometries, the axis of rotation varies as frequency
that require a nearby surface [21–24] and those that can swim
is changed while magnetic field strength stays constant [35].
in bulk fluids [25–35]. In this paper, we focus on magnetically
Previous studies have observed these changes in rotation axis
actuated microswimmers in bulk fluids which can be propelled
for magnetically rotated helices as frequency is varied [37–39].
when rotated by an external magnetic field [28–35].
The change in rotation axis leads to nonlinear velocity-
These magnetic swimmers are one of the most promising
frequency response and hinders ease of control. The rotation
types of microrobotic systems under current investigation.
axis can be characterized by a “precession angle” (β), which is
Compared to other swimmers, magnetic swimmers have
the angle the axis of rotation makes with the longitudinal heli-
advantages for in vivo use [36], since they do not require a
cal axis (Fig. 1). For β = 0◦ , the swimmer rotates as designed
potentially toxic chemical fuel, and magnetic fields permeate
around the helical axis, while for β = 90◦ the swimmer rotates
through tissue with little attenuation. Furthermore, swimmers
about an axis perpendicular to the helical axis, which has been
rotated via a magnetic torque produced by field-dipole interac-
called “tumbling.” Rotation about intermediate axes with 0◦ <
tions can be scaled down to micro- and nanoscales unlike those
β < 90◦ has been called “wobbling.” For example, Ghosh et al.
which rely on magnetic forces produced by field gradient-
[39] observed that helices magnetized with moments which
dipole interactions. Although nonhelical magnetically rotated
were not perpendicular to the helical axis tumble at the lowest
swimmers have been reported [35], nearly all such swimmers
frequencies and show little forward propulsion. As frequency
so far have been helical in geometry [28–30,32,34], mimicking
is increased past a critical frequency, the tumbling transitions to
the rigid helical flagella of bacteria. Thus, in this paper we
wobbling and the precession angle decreases while the velocity
focus on the swimming properties of helical geometries.
increases nonlinearly as frequency is further increased. The
These swimmers are actuated by a magnetic field of
precession angle approaches zero and the swimming velocity
magnitude H , which has a time-dependent direction that
peaks at the so-called step-out frequency. Beyond the step-out
rotates with angular velocity ω. In typical experiments, the
frequency, steady rotation is no longer possible since the
magnitude of the field is kept constant, while the rotation
magnetic torque is not large enough to rotate the swimmer
direction ω̂ and frequency ω = |ω| are varied. The direction
as fast as the field, and the swimming velocity decreases.
The tumbling and wobbling of magnetically rotated helices
has been theoretically examined in a number of previous
*
hfu@unr.edu studies. Man and Lauga [40] showed that during wobbling

1539-3755/2015/91(4)/043011(13) 043011-1 ©2015 American Physical Society

HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

as the aspect ratio decreases. We also compare swimming

speeds predicted by our theory to those reported for helical
swimmers in the literature and find good agreement. Finally,
in the context of these ideal directions of magnetization,
we investigate which helical geometries minimize wobbling
and maximize velocities. Previous work investigating optimal
shapes for helical swimmers [44] assumed that the rotation is
FIG. 1. (Color online) Rotation of the helix by angular velocity along the helical axis; here we incorporate the full wobbling
 can be characterized by the precession angle β. dynamics of helices rotated by magnetic fields.
In this work we use both resistive force theory [45] as
well as the method of regularized Stokeslets [46] to calculate
the precession angle scales as inverse frequency using nu- mobility matrices for helical geometries. The resistive force
merics and asymptotics of nearly straight helices. Ghosh theory is useful for obtaining analytical results which allow
et al. [39,41] and Morozov and Leshansky [42] investi- exploration of varying helical geometries, while the method
gated the transition of stability from tumbling to wobbling of regularized Stokeslets allows quantitative comparison with
behavior as frequency increases by treating the rotational experiments. Resistive force theory is commonly used to
dynamics as that of ellipsoids numerically and analytically, investigate both artificial microswimmers as well as biological
respectively. microswimmers, and by comparing the results of resistive
Two studies have explored what types of magnetization force theory and the method of regularized Stokeslets, we also
are optimal for helical microswimmers. Recently, Peters et al. show how thin the helix filament should be for the resistive
[37] proposed that wobbling can be eliminated by fabricating force theory approximation to hold. Besides its application to
superparamagnetic helical microswimmers in a way such that the helical swimmers, the work reported here also provides
the paramagnetic easy axis is perpendicular to the helical axis. quantitative validation of the modeling technique used. In [35]
They showed that microswimmers with such magnetization and [43] we reported the technique as a method applicable
had much reduced wobbling compared to microswimmers to generic geometries, providing qualitative comparison to
with random easy axis, as well as improved swimming observations in [35], but did not quantitatively compare it to
characteristics. In addition, in a theoretical examination of the experimental results. The agreement between our model and
rotational and swimming behavior of helices with permanent experimental swimming speeds for helices lends support to the
magnetic dipole, Morozov and Leshansky [42] concluded use of our model for a wide variety of geometries, including
that for such microswimmers, it is optimal to have the achiral geometries [35]. The potential insights allowed by our
magnetic dipole perpendicular to the helical axis to minimize technique are demonstrated by investigations of stability in
wobbling. varying magnetic field conditions [43], as well as the rotational
In previous publications [35,43], we have described a dynamics of helices reported here. Our model may be useful
modeling method for rigid magnetically rotated swimmers that for designing effective microswimmers from a much larger
is applicable for arbitrary geometries. The method identifies geometry space than helical geometries alone. Furthermore,
stable steady rotating orbits of the swimmer for given experi- the conclusion that a moment perpendicular to the principal
mental conditions in terms of the rotation axis of the swimmer axis of the rotational mobility matrix leads to linear velocity-
and the orientation of the swimmer relative to the magnetic frequency response applies to nonhelical geometries, which
field. Here we apply our techniques to investigate how the may be useful for the general design of microswimmers.
magnetization direction and geometry of helical swimmers
can be designed in order to control the rotation axis, reduce II. MODEL AND METHOD
wobbling, and produce linear velocity-frequency response,
which may enable simplified control, especially at the low We follow the general framework for modeling the ro-
speeds required for precise position control. tational propulsion of rigid bodies described in Ref. [43].
In the previous work on the rotational dynamics of helical Here we summarize the essential features for the present
microswimmers mentioned above [39,41,42], the rotational study; for more details, please see Ref. [43]. In the zero
dynamics was approximated by the dynamics of an ellipsoid or Reynolds number limit appropriate for microswimmers, the
rod, ignoring chirality and nonaxisymmetry of the helix. Here instantaneous velocity (v) and angular velocity () applied to
we treat the rotational dynamics of a truly helical geometry. a rigid body are linearly related to the external force (F) and
We investigate theoretically whether the direction of magne- torque (N) through a 6 × 6 mobility matrix,
tization can be used to guarantee wobble-free rotation about     
v K C F
the helical axis and find that it cannot. We show that although = . (1)
 CT M N
no moment can completely eliminate wobbling, there are ideal
directions of magnetization that are optimal in the sense that In the above, K, M, and C are 3 × 3 submatrices that relate
they can guarantee linear velocity-frequency relationships. translations to forces, rotations to torques, and translations to
For helices with large aspect ratios [(helical length)/(helical torques, respectively. According to this definition, K and M
diameter)], the ideal magnetization directions are close to are positive definite.
those suggested previously (i.e., perpendicular to the helical In typical experiments, a uniform field interacts with a
axis) by studies using the ellipsoidal approximation, and we magnetic dipole to produce zero external force and a net torque
provide a comparison to illustrate how the difference increases N = (m × H), so the instantaneous angular velocity of the

043011-2
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

constant direction and increases linearly with frequency, and so

Eq. (6) implies that Vs increases linearly with frequency. Thus,
maintaining a constant rotation axis for different frequencies
can be viewed as a design goal for helical swimmers.
One way to achieve a constant rotation axis is to increase
the magnetic field strength proportionally to the frequency as
the frequency of the rotation is increased [35]. In this case, it
should be clear that the magnetic field direction and rotation
direction satisfying Eq. (2) remain fixed while both the field
magnitude and frequency increase. However, impractically
FIG. 2. (Color online) (a) Helix with two turns, pitch P , and large magnetic fields may be required at higher frequencies, so
radius R. The helical axis is along x̂, and the helix is symmetric we search for other ways to achieve linear velocity-frequency
about a 180◦ rotation about ŷ. The principal axes of the rotational response.
mobility matrix M are ê1 , ê2 , and ê3 . (b) Angles φ and θ used for Steady solutions can be evaluated for stability by evaluating
expressing the magnetic field direction Ĥ in spherical coordinates. the eigenvalues of the matrix Q [43],
As specified in text, the angles may be used relative to the symmetry  
Qin = Mij j kl mk lmn Hm − ij n Mj l lmo mm Ho . (7)
(x,y,z) axes or the principal (1,2,3) axes.
If the real parts of the eigenvalues are all negative, then the
swimmer is solution is stable; otherwise, it is unstable.

 = M(m × H). (2) III. MOBILITY MATRIX FOR A HELIX

A steady solution occurs when the angular velocity of Consider a helix with centerline specified by
the swimmer is equal to the angular velocity of the rotating
magnetic field, so that in a body-fixed frame corotating with r(ξ ) = ξ x̂ + R cos(2πχξ/P )ŷ + R sin(2πχξ/P )ẑ, (8)
the swimmer, the magnetic field and hence angular velocity
where P is the helical pitch, R is the helical radius, ξ varies
are constant [47]. Thus, a steady solution obeys
from −nP /2 to nP /2, and n is the number of turns [see
 = ω = M(m × H), (3) Fig. 2(a)]. The parameter χ = ±1 controls the chirality of
the helix. Note that the arclength s = ξ/ cos α, where α =
and finding steady solutions amounts to identifying pairs of H arctan(2π R/P ) is the pitch angle.
and ω in the body-fixed frame that satisfy Eq. (3). In these coordinates, the helix is unchanged by the sym-
In many experiments, the magnetic field is rotated in a plane metry operation of a 180◦ rotation about the ŷ axis. Applying
perpendicular to its rotation axis. In this work, we restrict the symmetry arguments described in Chap. 5 of Happel and
ourselves to this condition unless explicitly stated otherwise. Brenner [48], the mobility matrix takes the form
For such a field, the steady solutions can be found by finding ⎛ ⎞
the magnetic field directions which satisfy Kxx 0 Kxz
K=⎝ 0 Kyy 0 ⎠, (9)
0 = ĤT M(m × Ĥ). (4) Kxz 0 Kzz
For example, the direction of the field can be specified using ⎛ ⎞
spherical coordinates defined in Fig. 2(b), Mx 0 Mxz
M=⎝ 0 My 0 ⎠, (10)
Ĥ = (cos θ, sin θ cos φ, sin θ sin φ) , (5) Mxz 0 Mz
in which case Eq. (4) provides a constraint on the angles (θ,φ). ⎛ ⎞
Cxx 0 Cxz
Once the magnetic field direction for a steady solution C=⎝ 0 Cyy 0 ⎠. (11)
is identified, the instantaneous swimming velocity can be Czx 0 Czz
calculated from Eq. (1) as v = CN = C(m × H), which is
also constant in the body-fixed frame. The average swimming In this study, we calculate the mobility matrix for a helix
velocity over a rotation period is the component of the using resistive force theory [45], which is often applied to
instantaneous swimming velocity along the rotation direction, slender objects such as helical filaments. Although resistive
force theory is not always quantitatively accurate [49], it is a
Vs = v · .
ˆ (6) convenient way to capture the qualitative dependence of the
Note that while the instantaneous velocity, which has been hydrodynamic mobility matrix on the helical geometry using
used by others as a measure of swimming speed, depends analytic expressions. Consider a helix moving due to external
on the origin chosen for the body-fixed frame, the swimming forces and torques in a quiescent fluid. In resistive force theory,
velocity of Eq. (6) is independent of the choice of origin. the force per unit length on a segment of the helix is
One way a linear relationship between swimming velocity
f(ξ ) = [ζ n̂n̂ + ζ⊥ (1 − n̂n̂)]vrel (ξ ), (12)
and frequency can be achieved is if the rotation axis remains
constant as the frequency is altered. In that case, Eq. (2) implies where vrel (ξ ) is the velocity of the segment relative to the
that the torque m × H also has constant direction and increases quiescent background, n̂(ξ ) is the tangent to the helix segment,
in magnitude linearly with the frequency. Therefore, v also has and ζ and ζ⊥ are the resistive force theory coefficients. The

043011-3
HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

ratio ζ⊥ /ζ depends only logarithmically on the thickness of frequency increases, H will rotate in the plane to make a larger
the helix relative to its pitch, so in this work, we keep the ratio angle with m and increase the torque. However, since the
constant (ζ⊥ /ζ = 2) for our qualitative investigations of the plane is perpendicular to (Rxx x̂ + Rzx ẑ) and not  = x̂,
effects of varying helical geometry. the angle between H and  must change as the frequency
The total force F and torque N on the helix are changes. In particular, for almost all frequencies H will not
nP /2 be perpendicular to . Thus, to achieve a rotation about the
dξ
F= f(ξ ), (13) helical axis requires a careful tuning of the angle between the
−nP /2 cos α field and its rotation axis as the frequency changes, which may
nP /2 be difficult to implement experimentally.
dξ
N= r(ξ ) × f(ξ ). (14) The above argument applies to any moment in the plane
−nP /2 cos α perpendicular to the torque. Of those, only one is also
If the helix undergoes rigid body motions with translational perpendicular to the helical axis, m = mŷ. Thus, for a helical
velocity v and angular velocity , then vrel (ξ ) = v + × geometry, in contrast to what is suggested by the ellipsoidal
r(ξ ). Calculating the total force and torque when v and  approximation, having a moment perpendicular to the helical
take values along Cartesian directions yields the elements of axis does not lead to rotation about the helical axis for fields
the resistance matrix, which is the inverse of the mobility perpendicular to rotation. Is there another strategy that could
matrix. In Appendix A we list the elements of the resistance be employed for easier control? Below, we show that although
matrix. The form of the matrix agrees with that predicted by rotation about the helical axis is not easily achievable, one
the symmetry analysis. can still choose a moment to obtain linear velocity-frequency
response.
IV. MAGNETIZATION PERPENDICULAR TO THE
HELICAL AXIS DOES NOT LEAD TO ROTATION V. ROTATION ABOUT THE PRINCIPAL AXIS OF M
ABOUT THE HELICAL AXIS
Although the rotational mobility submatrix M is not
Here we examine whether a permanent magnetic dipole diagonal when referred to the symmetry axes x̂, ŷ, ẑ, it is
oriented perpendicularly to the helical axis (i.e., in the y-z diagonal when referred to the principal axes ê1 , ê2 , ê3 . Call
plane), similar to that proposed by Peters et al. [37] and the respective eigenvalues M1 , M2 , and M3 , which are all
Morozov and Leshansky [42], leads to rotation about the positive. For typical helices with large aspect ratio, Mxz is small
helical axis ( = x̂) and hence no wobbling. compared to Mxx and Mzz and one of the eigenvectors of M
To do so, consider what moments and magnetic fields are will be close to x̂. Let ê1 be the principal axis closest to x̂. As the
needed to yield a rotation axis in the x direction. The inverse aspect ratio decreases, ê1 increasingly deviates from x̂. Below,
of M has the same form as M; call its nonzero elements we show that if the moment is chosen to lie perpendicular
⎛ ⎞ to ê1 , then the rotation axis will be along ê1 , independent
Rxx 0 Rzx
M−1 = ⎝ 0 Ryy 0 ⎠. (15) of frequency, and there will be a linear velocity-frequency
Rzx 0 Rzz response.
First, consider how one could obtain rotation along
In order to achieve a rotation x̂, Eq. (2) requires that the 1 direction ( = ê1 ). Repeating the argument
the torque m × H = (Rxx x̂ + Rzx ẑ). Note that the helical from the previous section but in the principal axis frame, since
geometry leads to off-diagonal elements Rzx so that the torque the mobility matrix is now diagonal the torque is in the 1
has a z as well as an x component. To satisfy this torque, direction, and m and H lie in the 2-3 plane. In this case, the
both m and H must lie in a plane perpendicular to (Rxx x̂ + 2-3 plane is always perpendicular to the rotation axis, so no
Rzx ẑ) (see Fig. 3). Consider what happens as the frequency of tuning of the angle between the field and the rotation axis is
rotation changes. For a given moment m in that plane, at small necessary as frequency is changed.
frequencies the field H will point in a direction close to m. As To be more specific, consider a magnetic moment in the
2-3 plane, m = m2 ê2 + m3 ê3 . Specify the field direction using
spherical coordinates as in Eq. (5), referred to the principal
axes (i.e., θ is the angle between H and the 1 axis). Then,
substituting into Eq. (4), the field directions giving steady
solutions satisfy the constraint

0 = sin θ cos θ [(M1 − M2 )m3 cos φ − (M1 − M3 )m2 sin φ] .

(16)

There are two families of solutions to the above constraint

equation.
FIG. 3. (Color online) For a rotation axis along the helical axis For the first family of solutions, θ = π/2 and H is in the
(x̂), the torque N is not in the x direction, and the moment m and field 2-3 plane (H = H2 ê2 + H3 ê3 ), leading to
H lie in a plane perpendicular to the torque. As the rotation frequency
changes, the angle H shifts to make a different angle γ with m.  = M1 (m2 H3 − m3 H2 )eˆ1 . (17)

043011-4
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

This is the solution discussed two paragraphs above, and is The second family of solutions is specified by
similar to the “propulsive” or “wobbling” solutions found (M1 − M2 )m3
under the ellipsoidal approximation employed by Ghosh tan φ = (18)
et al. [39,41] and Morozov and Leshansky [42,50]. Since the (M1 − M3 )m2
rotation axis is always in the 1 direction, it will have a linear for any value of θ . [Note that this includes solutions to Eq. (16)
velocity-frequency response. This solution exists as long as with θ = 0.] Referred to the principal axes, the corresponding
ω < M1 mH , the step-out frequency. field directions are given by

⎡ ⎤
(M1 − M3 )m2 (M1 − M2 )m3
Ĥ = ⎣cos θ, ± sin θ , ± sin θ ⎦, (19)
(M1 − M2 )2 m23 + (M1 − M3 )2 m22 (M1 − M2 )2 m23 + (M1 − M3 )2 m22

⎡ ⎤
M1 m2 m3 (M2 − M3 ) sin θ
 = H ⎣∓ ,M2 m3 cos θ, −m2 M3 cos θ ⎦ . (20)
(M1 − M2 )2 m23 − (M1 − M3 )2 m22

Vs can be found by direct substitution of Ĥ and  into Eq. (6), but the expression is not illuminating and too unwieldy to write
explicitly here. As we will see using a specific example below, this solution corresponds to “tumbling” rotation.
Note that although any moment in the 2-3 plane has a solution with constant rotation axis along the first principal direction, it
may be hard to experimentally realize such moments, since aligning swimmers with the principal, as opposed to symmetry, axes
may be difficult. However, ê2 = ŷ is perpendicular to both the symmetry and principal axis, and may be feasible to achieve during
fabrication. For example, if helices made by glancing angle deposition [30] are magnetized while still attached to substrate,
identification of the 180◦ symmetry axis corresponds to picking a particular direction parallel to the substrate. Therefore, in some
sense magnetization along ê2 = ŷ is the most practical option and we investigate it in more detail below.

Moment along the 2 direction

For the case of a moment m = m2 ê2 along the 2 (or equivalently y) direction, the above two families of solutions simplify to

 = M1 m2 H3 ê1 = H M1 m2 sin φ ê1 , (21)

H = H (cos φ ê2 + sin φ ê3 ), (22)

Vs = |m2 H3 |C11 = ωC11 /M1 , (23)

for the first family and

 = −M3 m2 H1 ê3 = −H M3 m2 cos θ ê3 , (24)

H = H (cos θ ê1 ± sin θ ê2 ), (25)

Vs = |m2 H1 |C33 = ωC33 /M3 , (26)

for the second family. It should be clear now that the since the second family rotates about the third principal axis, it is rotating
about an axis which is nearly perpendicular to the helical axis and represents the “tumbling” solution. The first family has two
solutions with opposite signs of  for 0 < ω < |H M1 m2 | (corresponding to φ and φ + π ), while the second family has two
solutions for 0 < ω < |H M3 m2 | (corresponding to the ±).
Both families are steady solutions, but to determine which solutions are realized experimentally we must evaluate the stability.
For the first family, Q in Eq. (7) is

⎛ ⎞
−m2 M1 cos φ 0 0
Q=H⎝ 0 0 m2 M1 sin φ ⎠ . (27)
0 −m2 (M1 − M3 ) sin φ −m2 M3 cos φ
√
The eigenvalues are λ1 = −M1 m2 cos φ and λ± = (A ± A2 + 4BC)/2, with A = −m2 M3 cos φ, B = M1 m2 sin φ, and C =
−m2 (M1 − M3 ) sin φ. For all helices we have examined, M1 > M3 . Thus, BC = −m22 M1 (M1 − M3 ) sin2 φ < 0 and the real
parts of λ± are both negative if and only if A = −m2 M3 cos φ < 0. Therefore, stability occurs for −π/2 < φ < π/2 if m2 > 0
but for π/2 < φ < 3π/2 if m2 < 0, and, in general, a stable solution exists for the whole range of ω where there are steady
solutions.

043011-5
HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

For the second family, Q is

⎛ ⎞
∓m2 M1 sin θ m2 (M1 − M3 ) cos θ 0
Q = H ⎝ m2 M3 cos θ 0 0 ⎠. (28)
0 0 ∓m2 M3 sin θ

√ eigenvalues are λ3 = ∓m2 M3 sin θ , and λ± = (A ±

The To conclude this section, we have shown that a moment
A2 + 4BC)/2, where now A = ∓m2 M1 sin θ , B = perpendicular to the principal axis rather than helical axis
m2 M3 cos θ , and C = m2 (M1 − M3 ) cos θ . In this case, BC = leads to a linear velocity-frequency response. Although the
m22 (M1 − M3 ) cos2 θ > 0, so that one of the eigenvalues λ± is resulting rotation axis (along ê1 ) is not along the helical axis,
always positive and the other negative. Thus, the second family for many helices it is relatively close. The dependence of
of solutions is always unstable. precession angle on helical geometry is explored in more detail
In Fig. 4 we plot the swimming velocity versus the fre- in Sec. VIII. Finally, note that although we only analytically
quency for the first and second families of solutions, for a helix proved that the propulsion solution is stable and tumbling
with n = 2, P = 4R, and chirality χ = 1. In the plots we use solution is unstable for moment along the 2 direction, for all
the dimensionless frequency specified by ω̃ = ωζ R 3 /(mH ) helical geometries we have tested, we observed numerically
and the dimensionless velocity Ṽs = Vs ζ R 2 /(mH ). Note that only the propulsive solution is stable as long as the moment
that for a given swimmer, the nondimensionalization of Vs is perpendicular to the first principal axis.
that we use is not affected by changes in frequency [unlike
another possibility, Vs /(ωR)]; thus, our plots directly reflect
VI. COMPARISON TO EXPERIMENTAL
the dimensional velocity-frequency plots commonly presented
SWIMMING SPEEDS
for experimental swimmers. For this chirality C11 > 0 and
C33 < 0, so the stable first family is the one with positive To validate our results, we compare to experimental results
slope that exists over a larger range. for microswimmers. The examples of microscale swimming
For this geometry of helix (n = 2, P = 4R), we can also magnetic helices in the literature do not exactly correspond
characterize the rotation via the precession angle β (Fig. 1), with the scenario which we discuss here. Ghosh and Fischer
which is the angle the axis of rotation makes with the helical [30] and Ghosh et al. [39] examine permanently magnetized
axis (x̂). For the first family, this is just the angle between ê1 helical swimmers with attached heads (which therefore do not
and x̂, giving β = 4.9◦ , while for the second family, it is the have the symmetry of a helix). Other swimmers are soft fer-
angle between ê3 and x̂, giving β = 85◦ . romagnetic helices with heads (e.g., [29]). While Peters et al.
[37] examine a helical ribbon with no head, their swimmer is
paramagnetic. Nonetheless, we can make rough comparisons
between these and our calculations to ensure that our models
are reasonable. Because typically the magnetic moment of
these swimmers is not measured the most appropriate measure
for comparison is the slope of the velocity-frequency curve
when nearly linear behavior is observed. If the frequency is
measured in hertz (revolutions/second), the slope corresponds
to the distance traveled per revolution, which is a property
dependent only on the geometry of the helix in our model.
Therefore, in the following we calculate the slope for the
propulsion solution with rotation about the first principal axis
of M and compare to experimental results.
Ghosh et al. [39] report artificial swimmers that consist of
a spherical head attached to a helix, which has a filament
diameter which tapers at the end away from the head. In
their paper, they find that swimming speeds are best modeled
using an “effective pitch,” the amount of translation after one
revolution about the helical axis, of 0.08 pitches/revolution.
The effective pitch is independent of the moment direction and
is a property of geometry of the helix. We model the tail as
a helix with constant diameter (see Appendix B for details).
In Secs. V, VII, and VIII, we use resistive force theory to
FIG. 4. (Color online) The swimming velocity vs frequency for calculate mobility matrices. Although resistive force theory
the stable solutions of a rotating helix with two turns, P = 4R, χ = 1, is useful for obtaining analytical expressions that reflect the
and a moment along the y direction. The lower branch corresponds symmetry of helices and can reveal qualitative trends, it is
to rotation around an axis with precession angle β = 85◦ , while the not expected to be quantitatively accurate for experimental
upper branch corresponds to rotation around an axis with precession swimmers with thick filaments. We find that resistive force
angle β = 4.9◦ theory is only accurate for filament radii less than 0.035P ,

043011-6
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

which are much thinner than experimental microswimmers the helical axis, but not perpendicular to the principal axis.
(see Appendix B for details). Thus, we calculate the mobility We compare it to the case described in Sec. V to show
matrix using the method of regularized Stokeslets [46] (see how moments perpendicular to the helical axis can lead to
Appendix B for details). For our model helix, we calculate undesirable swimming properties.
a velocity-frequency slope of 0.09 pitches and a precession Returning to Eq. (4), but referred to the symmetry axes
angle of <1◦ , which means that the rotation axis is nearly the x̂,ŷ,ẑ, we substitute magnetic field directions specified in
same as the helical axis. Our calculation does not include the spherical coordinates by Eq. (5), where θ is the angle from
spherical head; its effect can be approximated by calculating the x direction. This yields
the increase in drag in the direction of the helical axis  
0 = −m3 cos φ sin θ (Mx − My ) cos θ + Mxz sin φ sin θ ,
when a head is included in the mobility matrix calculation
and decreasing the effective pitch by the same proportion. (29)
With this correction, the estimated velocity-frequency slope
which admits solutions for steady orbits in two families.
is 0.085 pitches/revolution, comparable to the experimental
The first family is specified by
effective pitch of 0.08 pitches/revolution. Peters et al. [37]
report superparamagnetic ribbons wound into a helix with no Mx − M y
tan θ = − (30)
head, which they can rotate nearly without wobble. Again, Mxz sin φ
for such a thick and noncircular cross section resistive force or
theory is not expected to be accurate, so we calculate the
[Mxz sin φ x̂ + (My − Mx ) cos φ ŷ + (My − Mx ) sin φ ẑ]
mobility matrix using the method of regularized Stokeslets (see H=  .
2 sin2 φ + (M − M )2
Mxz
Appendix B for geometry). The resulting velocity-frequency x y
slope is 2.64 μm/revolution, which is in good agreement (31)
with the slopes from Figs. 8 and 9 of Peters et al., which From this equation,  and Vs can be obtained by substitution
have slopes of 2.6–2.8 μm/revolution. In contrast, Ghosh into Eqs. (2) and (6), but the expressions are not illuminating
and Fischer [30] report a microswimmer with similar (but so we do not write them here. This solution corresponds to the
smaller) geometry to that of Ghosh et al. [39] and find that the previously observed wobbling solutions [39].
velocity corresponds to 0.64 pitches per revolution. However, The second family has φ = π/2 and H in the x-z plane,
our model for this helix (see Appendix B for details) leads
H = H (cos θ x̂ + sin θ ẑ), (32)
to a velocity-frequency slope about an order of magnitude
smaller. We cannot explain the discrepancy; but we note that  = My mz Hx ŷ = H My mz cos θ ŷ, (33)
in addition to being inconsistent with our calculation, the quite
large value of 0.64 pitches per revolution is also inconsistent Vs = Cyy |mz Hx | = ωCyy /My , (34)
with the effective pitch reported by Ghosh et al. [39], as
well as biological helical propulsion, which always involves a which represents a tumbling solution around the y axis with
significant amount of slip of the helix with respect to the fluid. linear velocity-frequency response.
Based on our numerics and the imprecision of microswimmer Performing a stability analysis on the two families of
geometries, we expect our results to be accurate within 15% solutions reveals that the tumbling solutions are stable at
(see Appendix B for details). Thus, the results for our model are the lowest frequencies up to a critical frequency and become
in good overall agreement with the results from two different unstable at frequencies above the critical frequency ωc . Below
research groups (Ghosh et al. [39] and Peters et al. [37]). the critical frequency, the first family of solutions is unstable,
but above it the first family of solutions is stable. This is
similar to the behavior of ellipsoids with magnetic moments
which are not perpendicular to the long axis in studies where
VII. COMPARISON TO MOMENT PERPENDICULAR
the geometry of a helix is approximated as an ellipsoid [41,42].
TO HELICAL AXIS
We can obtain an expression for the critical frequency by
In this section, we calculate the rotational and swimming examining the stability matrix Q of Eq. (7) for the second
dynamics for a moment along ẑ, which is perpendicular to (tumbling) family, which is

⎛ ⎞
−Mx sin θ 0 (Mx − My ) cos θ
Q = H mz ⎝ 0 −My sin θ 0 ⎠. (35)
My cos θ − Mxz sin θ 0 Mxz cos θ

The eigenvalues are λ1 = −My sin θ , which is less shown that if Mx > My (which holds for helices with aspect
 over the entire range of θ , and λ± = [(A +
than zero ratio larger than 1), then both inequalities are satisfied for
D) ± (A + D)2 − 4(AD − BC)]/2. Here A = −Mx sin θ , (mz My H )2
B = (Mx − My ) cos θ , C = My cos θ − Mxz sin θ , and D = ω2 < ωc2 = . (36)
Mxz cos θ . In order for the tumbling solution to be stable, both 1 + (Mx − My )2 /Mxz
2

of the λ± must also be negative, which requires that both In Fig. 5(a), we plot the (nondimensional) velocity vs
(A + D) < 0 and (AD − BC) > 0. Using Eq. (33), it can be frequency for the stable solution as a function of frequency,

043011-7
HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

FIG. 5. The (a) swimming velocity and (b) precession angle vs frequency for the stable solutions of a rotating helix with two turns, P = 4R,
and a moment along the z direction, which is perpendicular to the helical axis but not the first principal axis. The inset to (a) magnifies the
low-frequency regime, illustrating the transition from tumbling with rotation about the y axis at the smallest frequencies to wobbling rotation
with varying precession angle at frequencies above the critical frequency ωc . The inset to (b) shows a log-log plot of the precession angle vs
frequency. The straight line has slope −1, demonstrating the relation β ∼ ω−1 .

for a helix with n = 2, P = 4R. The inset shows the transition perpendicular to the helical axis did not have stable tumbling
from the tumbling solution at low frequencies to the wobbling solutions. Here we show that by taking into account the
solution above ωc . Note that both the existence of the helical geometry, even moments perpendicular to the helical
transition and the frequency dependence of the wobbling axis lead to tumbling-wobbling transitions and nonlinear
solutions lead to nonlinear dependence of the velocity on velocity-frequency response. In order to achieve the linear
frequency, which are disadvantageous for precise swimmer velocity-frequency response, one must instead target rotation
control, especially at low frequencies. At higher frequencies about the principal axis using a moment perpendicular to ê1 .
the velocity depends on the frequency nearly linearly. The In Fig. 5(b), we plot the precession angle β as a function of
form of the dependence is similar to that observed for helices frequency for the stable solution. Below the critical frequency,
by Ghosh et al. [39] and modeled for ellipsoids with moment the tumbling solution rotates around the y axis, so β = 90◦ .
nonperpendicular to the long axis by Ghosh et al. [41] and Above the critical frequency, the wobbling solution starts with
Morozov and Leshansky [42]. However, those previous studies β = 90◦ and decreases as frequency increases. Man and Lauga
had found that in the ellipsoidal approximation, any moment have shown that β ∼ ω−1 for asymptotically straight helices.

FIG. 6. The swimming velocity (a) and precession angle (b) vs frequency for the stable solutions of rotating helices with moment along the
z direction. From left to right [for (a), as measured by location of sketched helices], and bottom to top [for (b)], curves correspond to helices
with: two turns, P = 4R [same case as in Fig. 5]; two turns, P = 2R; one turn, P = 5R; one turn, P = 3R; one turn, P = 2R. Curves are
only plotted for frequencies below the step-out frequency. As the aspect ratio (length/diameter) of the helix decreases, a greater portion of
frequencies below step-out exhibit tumbling rotation.

043011-8
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

In the inset of Fig. 5(b), we plot the precession angle as a

function of frequency in a log-log plot to show that for a
range of frequencies our precession angle has the same ω−1
dependence (straight line).
Figure 5 shows that for a helix with n = 2 and P = 4R,
the critical frequency is relatively small compared the the
step-out frequency, so that the nonlinear response regime is
quite small. Is the nonlinear response regime always small?
To investigate this, in Fig. 6 we plot the velocity-frequency
response for helices with a number of different geometries. The
general trend is that as the aspect ratio [(helical length)/(helical
diameter)] decreases, the critical frequency increases and the
nonlinear regime grows in size. For a helix with n = 1 and
P = 2R, ωc is about a third of the step-out frequency, and
the nonlinear regime is a significant portion of the frequency
range below step-out. Thus, results from the ellipsoidal
approximations, which predict that rotation about the helical
axis can be achieved for moments perpendicular to the helical
axis, fail strongly when the aspect ratio of helices is no longer
large.

VIII. DEPENDENCE ON HELICAL GEOMETRY

FIG. 7. (Color online) The precession angle of a helix with
In this section, we continue to investigate the influence moment along the y direction, perpendicular to the first principal
of helical geometry on swimming properties. We return to axis, as a function of ratio of helical pitch and radius (P /R), for
our proposed moment m = m2 ŷ, which is perpendicular to helices with different numbers of turns: one turn (red, thick solid
the principal axis ê1 and leads to rotation about ê1 and a line), two turns (orange, thick dashed line), four turns (green, thin
linear velocity-frequency response. For this moment direction, solid line), and eight turns (blue, thin dashed line).
we investigate which helical geometries will lead to (a) low
precession angles, (b) maximum velocity:frequency slopes, the velocity:frequency slope, since it is the product of the
and (c) greatest maximum velocity. velocity:frequency slope and the step-out frequency. The
In Fig. 7 we plot the precession angle of the swimmer as a choice of whether it is more important to optimize
function of helical pitch for various numbers of turns. Recall
that this precession angle is simply the angle between ê1 and
the x axis. A small precession angle may be desirable if one
wishes to precisely control the location of the tip of the helix
during swimming. Typically, the precession angle decreases
for larger aspect ratios. Specifically, in order to achieve a
precession angle β < 10◦ , for n  2 the pitch must satisfy
P > 2R, while for n = 1, the pitch must satisfy P > 8. These
results are in accord with the arguments made by Morozov and
Leshansky about the effect of helical geometry on rotational
dynamics modeled using ellipsoids [42]; here we quantify how
the helical geometry leads to a breakdown of the ellipsoidal
approximation.
In Fig. 8 we plot the slope of the velocity-frequency
response Ṽs /ω̃ of the swimmer as a function of helical
pitch for various numbers of turns. Recall that the dimen-
sionless frequency is ω̃ = ωζ R 3 /(mH ) and the dimension-
less velocity is Ṽs = Vs ζ R 2 /(mH ); thus, the dimensionless
slope is nondimensionalized by the helical radius R, i.e.,
Ṽs /ω̃ = (Vs /ω)/R. Therefore, Fig. 8 is best interpreted as
investigating the variation of the slope as the helical pitch
changes but the helical radius stays fixed. For a fixed number of
turns, the maximum slope occurs for a pitch around 8R–10R. FIG. 8. (Color online) The velocity:frequency ratio of a helix
As the number of turns increases, the slope increases, but with moment along the y direction, perpendicular to the first principal
saturates around n = 8. axis, as a function of ratio of helical pitch and radius (P /R), for helices
In Fig. 9(a) we plot the maximum velocity Ṽsmax of the with different numbers of turns: one turn (red, thick solid line), two
swimmer as a function of helical pitch for various numbers turns (orange, thick dashed line), four turns (green, thin solid line),
of turns. Note that the maximum velocity is distinct from and eight turns (blue, thin dashed line).

043011-9
HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

turns (n = 8) have the largest velocity:frequency slope in

Fig. 8 but the smallest maximum velocity in Fig. 9; the large
velocity:frequency slope is multiplied by a small step-out
frequency [Fig. 9(b)] to yield small maximum velocities.
Based on Fig. 9(a), the fastest swimmers would be obtained
for n = 1 and P ≈ 5R; however, these would have relatively
large precession angles (β ≈ 15◦ ). For n = 2 and P = 4R, the
maximum velocity is about half as much, but the precession
angle is reduced to β ≈ 5◦ .
Note that all the discussion of geometrical dependence in
this section uses the helical radius R to nondimensionalize the
velocities. Thus, the results should be interpreted as keeping
the helical radius R fixed while the pitch and number of turns
are varied. For the most common fabrication processes, it is
easiest to maintain constant helical radius R as we have done.
However, if one wished to design helices with varying R,
the results would be different since the scaling of our plotted
results with R would need to be taken into account.

IX. DISCUSSION
We investigated the rotational and swimming dynamics of
rigid helical microswimmers rotated by an external magnetic
field. We investigated how the rotation axis and dependence of
velocity on frequency is affected by the direction of magnetic
moment and geometry of the helix. We use resistive force
theory to obtain mobility matrices that capture the helical
geometry of the swimmers, going beyond approximations
which treat the rotational dynamics of such swimmers as that
of ellipsoids. A linear velocity-frequency response is desirable
for control of such microswimmers, and we show that this
can be achieved by choosing the magnetic moment to lie
perpendicular to the principal axis closest to the helical axis,
which results in a single stable branch of solutions which all
rotate about the principal axis. We also show that moments
which are perpendicular to the helical axis rather than the
principal axis lead to nonlinear velocity-frequency response
including a transition between low-frequency tumbling and
high-frequency wobbling dynamics. Finally, we explored the
dependence of swimming properties on helical geometry in the
context of our proposed moment perpendicular to the principal
axis. Precession angle, the slope of the velocity-frequency
FIG. 9. (Color online) (a) The maximum velocity of a helix with response, and maximum velocity were optimized for different
moment along the y direction, perpendicular to the first principal helical geometries.
axis, as a function of ratio of helical pitch and radius (P /R), for Our work used the resistive force theory to obtain qualita-
helices with different numbers of turns: one turn (red circles), two tively accurate mobility matrices for the helical swimmers.
turns (orange squares), four turns (green diamonds), and eight turns The resistive force theory is sufficient to investigate the
(blue triangles). (b) The step-out frequency of the same helices as a trends in behavior by providing a convenient way to calculate
function of P /R. the mobility matrices for arbitrary helical pitch and radius
and is sufficient to explore qualitative features of linear
velocity:frequency slope or maximum velocity depends on the velocity-frequency relationships. While resistive force theory
application. For a fixed number of turns, the maximum velocity is accurate for very slender helix filaments, we used a boundary
occurs for pitches around 4R–6R, at smaller pitch values than element method to obtain quantitatively accurate mobility
the maximum slope. This can be rationalized by observing matrices for the thicker helices fabricated in experiments
how the step-out frequency ω̃stepout behaves for different helical so far and obtained good agreement between our calculated
geometries, as plotted in Fig. 9(b). For a fixed number of turns, swimming speeds and experimental results.
the step-out frequency is highest for small pitches near R. Since We assumed a helical geometry, but if a nonhelical geom-
the maximum velocity is the product of the slope (Fig. 8) etry is used, the mobility matrix may be more complicated
and the step-out frequency [Fig. 9(b)], it is at intermediate than Eq. (1) due to a loss of the 180◦ rotation symmetry. For
pitch. Similarly, one can rationalize why swimmers with many example, although a helical microswimmer with a head or

043011-10
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

cargo [29,30] will not have the symmetries we used in our APPENDIX A: RESISTANCE AND MOBILITY MATRIX
analysis, this can be addressed in two ways. First, if a second FOR A HELIX
head is attached at the other end of the swimmer, the symmetry
Here we report the coefficients of the resistance matrix for
upon 180◦ rotation may be restored. Second, even if the
a helix with pitch P , radius R, and n turns calculated using
swimmer is not symmetric, one can identify the principal axes
resistive force theory as described in the main text. Due to
of its mobility matrix. In that case, the results of Sec. V apply
the symmetry, the resistance matrix D has the same nonzero
insofar as a moment perpendicular to a principal axis will admit
elements as the mobility matrix in the x̂, ŷ, ẑ bases,
steady rotational solutions about the principal axis for a range
of frequencies. In addition, rotation about the principal axis ⎛ ⎞
D11 0 D13 D14 0 D16
with the smallest rotational mobility eigenvalue will be stable. ⎜ 0 D22 0 0 D25 0 ⎟
Our work differs from that of Ghosh et al. [41] and Morozov ⎜ ⎟
⎜D 0 D33 D34 0 D36 ⎟
and Leshansky [42] in that they approximate the rotational D = ⎜ 13 ⎟. (A1)
⎜D14 0 D34 D44 0 D46 ⎟
dynamics as those of an magnetized ellipsoid. One conclusion ⎝ 0 D25 0 0 D55 0 ⎠
of their work is that a magnetization perpendicular to the D16 0 D36 D46 0 D66
helical axis can reduce wobbling and lead to rotation about
the helical axis. As shown in Secs. IV–VII, using the true For integer n, the elements above are specified by
helical geometry means that rotation about the helical axis is
not actually feasible, and instead rotation about the principal D11 = (ζ R)n(P̃ cos α + 2π ζ̃⊥ sin α), (A2)
axis, which differs from the helical axis, is a better target. In-
corporating the helical geometry, we find tumbling-wobbling D22 = −(ζ R)π n[(ζ̃⊥ − 1) sin α − 2ζ̃⊥ / sin α], (A3)
transitions and nonlinear velocity-frequency response even
when the moment is perpendicular to the helical axis. Although D33 = D22 , (A4)
the difference between principal axis and helical axis is small
for helices with large aspect ratio, which justifies the ellipsoidal D13 = 0, (A5)
approximation in those cases, the principal axis and helical axis
can be significantly different for helices with smaller aspect D44 = −(ζ R 3 )2π n[(ζ̃⊥ − 1) sin α − ζ̃⊥ / sin α], (A6)
ratios. Thus, the helical geometry alters the conclusions of
D55 = (ζ R 3 ) 12
1
n{n2 P̃ 3 ζ̃⊥ cos α + 12π ζ̃⊥ sin α
those earlier studies in a manner which may be significant
for the design of microswimmers. We propose that the most + [(15 + 2n2 π 2 ) − (9 − 2n2 π 2 )ζ̃⊥ ]P̃ cos α}, (A7)
convenient moment to target experimentally is one which is
perpendicular to both the first principal axis and the helical D66 = (ζ R 3 ) 12
1
n{n2 P̃ 3 ζ̃⊥ cos α + 12π ζ̃⊥ sin α
axis, rather than any direction perpendicular to the helical axis.
In this work we assumed a permanent magnetic dipole + [(2n2 π 2 − 3) + (9 + 2n2 π 2 )ζ̃⊥ ]P̃ cos α}, (A8)
and ignored paramagnetic response of the swimmers. Such
D46 = −(ζ R 3 )nP̃ χ [(ζ̃⊥ − 1) sin α − ζ̃⊥ / sin α], (A9)
response has been addressed by Morozov and Leshansky [50].
In that study, rotational dynamics are approximated as that
D14 = −(ζ R 2 )2π nχ (ζ̃⊥ − 1) cos(α), (A10)
of an ellipsoid, and we would expect that taking into account
the helical geometry may yield similar differences with the D25 = (ζ R 2 ) 23 nπ χ (ζ̃⊥ − 1) cos α, (A11)
ellipsoidal approximation that we have found in the case
of a permanent magnetic dipole. Note that in particular the D36 = (ζ R 2 ) 21 nπ χ (ζ̃⊥ − 1) cos α, (A12)
paramagnetic response would be expected to alter the linear
velocity-frequency relations found in this work. D16 = −(ζ R 2 )nP̃ (ζ̃⊥ − 1) cos α, (A13)
In this work we also ignored interactions with boundaries
and walls. Although many research groups have explored mag- D34 = 0. (A14)
netically rotated microswimmers which “roll” along surfaces,
the modeling techniques used here are not directly applicable Where P̃ = P /R and ζ̃⊥ = ζ⊥ /ζ . The mobility matrix is the
since the presence of the wall means that the steady solutions inverse of the resistance matrix. Our resistance matrix differs
found here do not exist; instead, there is a time-dependent from that reported in Man and Lauga [40], since, according
mobility matrix as the swimmer rotates relative to the wall. to their Eq. (1), their origin is located at one end of the helix,
Treatment of such cases would require integration of such while ours is located along the symmetry axis in the center of
unsteady rotational orbits, and we do not expect our statements the helix. They also include contributions to the moment from
about rotation axes to hold when interaction with surfaces and local moment densities which resistive force theory ignores.
boundaries becomes strong.
APPENDIX B: QUANTITATIVE CALCULATION
ACKNOWLEDGMENTS OF MOBILITY MATRICES
This work was supported by National Science Foundation In most of the paper, we used resistive force theory with the
Awards No. DMR-1307497 and No. CMMI-1435652 to H.F. ratio of perpendicular to parallel resistive force coefficients

043011-11
HENRY C. FU, MEHDI JABBARZADEH, AND FARSHAD MESHKATI PHYSICAL REVIEW E 91, 043011 (2015)

FIG. 10. (Color online) (a) Geometry used to model helix of

Ghosh et al. [39]. (b) Geometry used to model helix of Peters et al.
[37].

equal to 2, which is the expected value in the limit of very FIG. 11. (Color online) Convergence study for modeling helix
slender helical filaments. As the filament radius (a) becomes from [39]: velocity-frequency slope as a function of number of
thicker, one can adjust resistive force coefficients; for helices, regularized Stokeslets used in discretization.
the following formulas apply [45]:
2π μ for a = 0.128 μm and negative parallel coefficients for a =
ζ = , (B1) 0.19 μm.
ln(0.18P /a)
The helical ribbon of Peters et al. [37] was modeled as a
4π μ helix with R = 2.9 μm, P = 16 μm, three turns, and a cross
ζ⊥ = . (B2)
ln(0.18P /a) + 1/2 section with width 2.8 μm and depth 1.5 μm. The geometry
However, for the experimental geometries described below is shown in Fig. 10(b). This geometry was obtained from
(e.g., P = 0.91 μm, a = 0.19 μm), these formulas yield Fig. 3 of [37]. Based on the convergence study performed
negative ζ , implying that they are too thick to be adequately for the Ghosh et al. helix, the reported results are obtained
treated by resistive force theory. from discretizations involving 5832 regularized Stokeslets,
To calculate quantitatively accurate mobility matrices, we for which we expect discretization error of ≈5%. Finally, we
used the method of regularized Stokeslets [46,51]. Our group note that we observed that changes in geometry, particularly
has previously implemented the method [52], including to find in the length of the helix, could produce roughly propor-
mobility matrices for the modeling procedure employed in this tional changes in the velocity-frequency slope. Additionally,
paper [35,43], and details of the method can be found in those changing the depth and width of the Peters et al. helix to
references. 3.3 and 1.8 μm (10%–20% changes) reduced the swimming
Here we provide geometries and results of convergence speed by approximately 10%. Based on the imprecision of
studies for the helical microswimmers discussed in Sec. VI. We our measurements of experimental geometries, which also
use these to provide estimates for the error in our calculations do not form perfectly regular helices, we therefore expect
of velocity-frequency slopes. additional 5%–10% errors in our results. Combining with
The helical swimmer of Ghosh et al. [39] was modeled as the discretization error, we estimate (conservatively) that our
a helix with R = 0.16 μm, P = 0.91 μm, four turns, and a calculated velocity-frequency slopes are accurate to within
filament diameter of 0.38 μm, as shown in Fig. 10(a). This 15%.
geometry was obtained from Fig. 1 of [39]. We performed a
convergence study by discretizing the surface with varying
numbers of regularized Stokeslets (Fig. 11). The results
reported in the main text are for the largest number of elements,
12 134 regularized Stokeslets, but we see that even for ≈6000
Stokeslets there is <5% error.
Based on the convergence study above, we can also
investigate the accuracy of resistive force theory results for
the velocity:frequency slope. In Fig. 12 we compare results
from resistive force theory to the method of regularized
Stokeslets for the geometry above but for varying filament
radius. The number of discretization elements for the method
of regularized Stokeslets varies from 6974 to 15 544. For
filament radii a < 0.032 μm (a/P < 0.035), there is less
than 5% difference between the resistive force theory and
method of regularized Stokeslets. For resistive force theory, FIG. 12. (Color online) Comparison of velocity:frequency
the two largest filament radii plotted (a = 0.128 μm and slopes calculated using resistive force theory (blue circles) and
a = 0.19 μm) give unphysical negative results, which reflect method of regularized Stokeslets (black squares) as a function of
unphysical resistive force coefficients: Eqs. (B1) and (B2) filament thickness. The helix geometry is from [39], but with the
give parallel coefficients larger than perpendicular coefficients filament radius a altered.

043011-12
MAGNETIZATION DIRECTIONS AND GEOMETRIES OF . . . PHYSICAL REVIEW E 91, 043011 (2015)

[1] G. Dogangil, O. Ergeneman, J. Abbott, S. Pane, H. Hall, S. [26] W. Gao, D. Kagan, O. S. Pak, C. Clawson, S. Campuzano,
Muntwyler, and B. Nelson, IEEE/RSJ International Conference E. Chuluun-Erdene, E. Shipton, E. E. Fullerton, L. Zhang,
on Intelligent Robots and Systems, 2008 (IEEE, Piscataway, NJ, E. Lauga et al., Small 8, 460 (2012).
2008), pp. 1921–1926. [27] O. S. Pak, W. Gao, J. Wang, and E. Lauga, Soft Matter 7, 8169
[2] S. Fusco, G. Chatzipirpiridis, K. M. Sivaraman, O. Ergeneman, (2011).
B. J. Nelson, and S. Pan, Adv. Healthcare Mater. 2, 1037 (2013). [28] L. Zhang, E. Ruh, D. Grützmacher, L. Dong, D. J. Bell, B. J.
[3] A. Ferreira, J. Agnus, N. Chaillet, and J. M. Breguet, Nelson, and C. Schönenberger, Nano Lett. 6, 1311 (2006).
IEEE/ASME Trans. Mechatronics 9, 508 (2004). [29] S. Tottori, L. Zhang, F. Qiu, K. K. Krawczyk, A. Franco-
[4] H. Zhang, D. W. Hutmacher, F. Chollet, A. N. Poo, and E. Obregón, and B. J. Nelson, Adv. Mater. 24, 811 (2012).
Burdet, Macromol. Biosci. 5, 477 (2005). [30] A. Ghosh and P. Fischer, Nano Lett. 9, 2243 (2009).
[5] S. Kim, F. Qiu, S. Kim, A. Ghanbari, C. Moon, L. Zhang, B. J. [31] U. K. Cheang, D. Roy, J. H. Lee, and M. J. Kim, Appl. Phys.
Nelson, and H. Choi, Adv. Mater. 25, 5863 (2013). Lett. 97, 213704 (2010).
[6] X. Liu, K. Kim, Y. Zhang, and Y. Sun, Int. J. Rob. Res. 28, 1065 [32] F. Z. Temel and S. Yesilyurt, IEEE International Conference
(2009), http://ijr.sagepub.com/content/28/8/1065.full.pdf+html on Mechatronics (ICM), 2011 (IEEE, Piscataway, NJ, 2011),
[7] S. Martel, J.-B. Mathieu, O. Felfoul, A. Chanu, E. Aboussouan, pp. 342–347.
S. Tamaz, P. Pouponneau, L. Yahia, G. Beaudoin, G. Soulez, [33] P. Garstecki, P. Tierno, D. B. Weibel, F. Sagués, and G. M.
and M. Mankiewicz, Appl. Phys. Lett. 90, 114105 (2007). Whitesides, J. Phys.: Condens. Matter 21, 204110 (2009).
[8] M. S. Grady, M. A. Howard, J. A. Molloy, R. C. Ritter, E. G. [34] L. Zhang, J. J. Abbott, L. Dong, B. E. Kratochvil, D. Bell, and
Quate, and G. T. Gillies, Med. Phys. 17, 405 (1990). B. J. Nelson, Appl. Phys. Lett. 94, 064107 (2009).
[9] J.-B. Mathieu, G. Beaudoin, and S. Martel, IEEE Trans. Biomed. [35] M. Jabbarzadeh, Y. Hyon, and H. C. Fu, Phys. Rev. E 90, 043021
Eng. 53, 292 (2006). (2014).
[10] R. F. Ismagilov, A. Schwartz, N. Bowden, and G. M. Whitesides, [36] J. J. Abbott, K. E. Peyer, M. C. Lagomarsino, L. Zhang, L. Dong,
Angew. Chem. 114, 674 (2002). I. K. Kaliakatsos, and B. J. Nelson, Int. J. Rob. Res. 28, 1434
[11] W. F. Paxton, K. C. Kistler, C. C. Olmeda, A. Sen, S. K. St. (2009).
Angelo, Y. Cao, T. E. Mallouk, P. E. Lammert, and V. H. Crespi, [37] C. Peters, O. Ergeneman, B. J. Nelson, and C. Hierold,
J. Am. Chem. Soc. 126, 13424 (2004). IEEE 26th International Conference on Micro Electro Me-
[12] Y. Wang, R. M. Hernandez, D. J. Bartlett, J. M. Bingham, chanical Systems (MEMS), 2013 (IEEE, Piscataway, NJ, 2013),
T. R. Kline, A. Sen, and T. E. Mallouk, Langmuir 22, 10451 pp. 564–567.
(2006). [38] K. E. Peyer, L. Zhang, B. E. Kratochvil, and B. J. Nelson, IEEE
[13] W. Gao, M. D’Agostino, V. Garcia-Gradilla, J. Orozco, and International Conference on Robotics and Automation (ICRA),
J. Wang, Small 9, 467 (2013). 2010 (IEEE, Piscataway, NJ, 2010), p. 96.
[14] J. Orozco, V. Garcı́a-Gradilla, M. DAgostino, W. Gao, A. Cortes, [39] A. Ghosh, D. Paria, H. J. Singh, P. L. Venugopalan, and A.
and J. Wang, ACS Nano 7, 818 (2012). Ghosh, Phys. Rev. E 86, 031401 (2012).
[15] K. M. Manesh, M. Cardona, R. Yuan, M. Clark, D. Kagan, [40] Y. Man and E. Lauga, Phys. Fluids 25, 071904 (2013).
S. Balasubramanian, and J. Wang, ACS Nano 4, 1799 (2010). [41] A. Ghosh, P. Mandal, S. Karmakar, and A. Ghosh, Phys. Chem.
[16] A. A. Solovev, Y. Mei, E. Bermúdez Ureña, G. Huang, and Chem. Phys. 15, 10817 (2013).
O. G. Schmidt, Small 5, 1688 (2009). [42] K. I. Morozov and A. M. Leshansky, Nanoscale 6, 1580 (2014).
[17] J. Gibbs and Y.-P. Zhao, Appl. Phys. Lett. 94, 163104 (2009). [43] F. Meshkati and H. C. Fu, Phys. Rev. E 90, 063006 (2014).
[18] T. R. Kline, W. F. Paxton, T. E. Mallouk, and A. Sen, Angew. [44] E. E. Keaveny, S. W. Walker, and M. J. Shelley, Nano Lett. 13,
Chem. 117, 754 (2005). 531 (2013).
[19] E. B. Steager, M. S. Sakar, D. H. Kim, V. Kumar, G. J. [45] S. Childress, Mechanics of Swimming and Flying (Cambridge
Pappas, and M. J. Kim, J. Micromech. Microeng. 21, 035001 University Press, Cambridge, UK, 1981).
(2011). [46] R. Cortez, SIAM J. Sci. Comput. 23, 1204 (2001).
[20] D. L. Fan, F. Q. Zhu, R. C. Cammarata, and C. L. Chien, Nano [47] Our definition of steady solution excludes the solutions with
Today 6, 339 (2011). phase slip discussed by Ghosh et al. [41] and Morozov and
[21] C. E. Sing, L. Schmid, M. F. Schneider, T. Franke, and Leshansky [42] .
A. Alexander-Katz, Proc. Natl. Acad. Sci. USA 107, 535 [48] J. Happel and H. Brenner, Low Reynolds Number Hydrodynam-
(2010). ics: With Special Applications to Particulate Media (Springer,
[22] P. Tierno, O. Güell, F. Sagués, R. Golestanian, and Berlin, 1983), Vol. 1.
I. Pagonabarraga, Phys. Rev. E 81, 011402 (2010). [49] B. Rodenborn, C.-H. Chen, H. L. Swinney, B. Liu, and H. P.
[23] L. Zhang, T. Petit, Y. Lu, B. E. Kratochvil, K. E. Peyer, R. Pei, Zhang, Proc. Natl. Acad. Sci. USA 110, E338 (2013).
J. Lou, and B. J. Nelson, ACS Nano 4, 6228 (2010). [50] K. Morozov and A. Leshansky, Nanoscale 6, 12142 (2014).
[24] W. Xi, A. A. Solovev, A. N. Ananth, D. H. Gracias, S. Sanchez, [51] R. Cortez, L. Fauci, and A. Medovikov, Phys. Fluids 17, 031504
and O. G. Schmidt, Nanoscale 5, 1294 (2013). (2005).
[25] R. Dreyfus, J. Baudry, M. L. Roper, M. Fermigier, H. A. Stone, [52] Y. Hyon, Marcos, T. R. Powers, R. Stocker, and H. C. Fu,
and J. Bibette, Nature (London) 437, 862 (2005). J. Fluid Mech. 705, 58 (2012).

043011-13