Orientational dynamics and rheology of active suspensions in weakly viscoelastic flows

Akash Choudhary1 ,∗ Sankalp Nambiar2 , and Holger Stark1†

1
Institute of Theoretical Physics, Technische Universität Berlin, Hardenbergstr. 36, 10623 Berlin, Germany
2
Nordita, KTH Royal Institute of Technology and Stockholm University, Stockholm 10691, Sweden

Abstract. Microswimmer suspensions are constantly in non-equilibrium and exhibit macroscale

properties that are in stark contrast to passive suspensions. Motivated by ubiquitous microbial sys-
tems suspended in biological fluids, we analyse the rheological response of a suspension of elongated
microswimmers to a steady shear flow in a weakly viscoelastic fluid. At the individual level, we
find that the viscoelastic stresses generated by activity substantially modify the Jeffery orbits well-
known from Newtonian fluids. The orientational dynamics depends on the swimmer type and, in
particular, microswimmers can resist flow-induced rotation and align at an angle with the flow. To
arXiv:2303.15241v1 [cond-mat.soft] 27 Mar 2023

analyze its impact on the bulk rheological response, we study a dilute ensemble of microswimmers in
the presence of stochastic noise due to tumbling and rotational diffusion. Compared to Newtonian
fluids, activity and its elastic response in polymeric fluids alter the orientational distribution and
substantially amplify the swimmer-induced viscosity. This suggests that pusher suspensions reach
the regime of superfluidity at lower volume fractions.

INTRODUCTION even down to the ‘superfluidic’ limit [21]. The mecha-

Systems of particulate matter suspended in fluids
are prevalent in numerous natural and industrial pro-
cesses. Active suspensions are particulate systems that
are driven out of equilibrium by converting chemical
energy or fuel into mechanical work to achieve self-
propulsion [1, 2]. Motile microorganisms are the pro-
totypical example of active motion that is generated
via metachronal actuation of hairlike cilia (Paramecium,
Volvox ), by whipping cell-attached flagellar appendages
(spermatozoa, algae), or by rotating a bundle of helical
flagella (bacteria) [3]. Microswimmers navigate through
their environment by sensing or interacting with gradi-
ents in hydrodynamic, chemical, thermal, and light fields;
a strategy known as taxis [4–10]. A particular example
is rheotaxis, where microswimmers experience hydrody-
namic gradients and swim against the flow, which is rele-
vant for biofilm formation and reproduction [2, 8, 11, 12].
Pathogenic microswimmers when infiltrating human
and animal bodies have to pass through mucus linings
that lubricate and protect our respiratory tracks, eyes,
urogenital and gastrointestinal systems [13]. The pres-
ence of mucin fibres (typically 3-10 nm in length [14])
and DNA makes these mucus linings viscoelastic and
shear thinning [15]. The linings are often subjected to
shearing motion, for instance, during blinking, cough-
ing, reproduction, and continuos mucociliary clearance
in respiratory systems. In such microbiological flows, the
non-Newtonian fluid properties can alter the swimmer’s
rheotactic behaviour [16, 17].
In this article we study the bulk rheology of mi-
croswimmers in viscoelastic flows. For Newtonian flu-
ids, several rheological experiments on suspensions of
extensile swimmers like E.coli have shown that activ-
ity can drastically reduce the effective viscosity [18–20],
∗ a.choudhary@campus.tu-berlin.de
† holger.stark@tu-berlin.de
nism, first outlined in Ref. [22] and elaborated further in
Refs. [23, 24], is as follows. The orientations of elongated
swimmers in shear flow follow periodic Jeffery orbits [25]
and are also subject to thermal noise. This competition
yields a mean orientation that points in the extensional
quadrant of applied shear flow. With such an orientation
extensile microswimmers (pushers) support the applied
shear flow and thereby reduce the effective shear vis-
cosity, whereas contractile swimmers like C. reinhardtii
(pullers) resist it and thus increase viscosity [26].
Since almost all biological fluids are non-Newtonian,
there has been a recent interest towards developing theo-
retical frameworks that capture the individual and collec-
tive dynamics of microswimmers in complex fluids [27–
29]. For example, experiments have shown reduced tum-
bling and increased persistence lengths of bacteria in
polymeric fluids [30–32]. Viscoelasticity can also initi-
ate spatiotemporal order in active suspensions. For ex-
ample, a recent study found that DNA polymers trig-
ger oscillatory vortices in confined suspension of E.coli
[33]. However, determining the effective shear viscosity
of an active suspension in a viscoelastic fluid has been an
uncharted territory because of its complexity: the elas-
tic relaxation of non-Newtonian fluids and their shear-
thinning/thickening property.
To gain better insights into biological and artificial
motility in biological fluids, this communication inves-
tigates the role of non-linear polymeric stresses in non-
Newtonian fluids. Specifically, this non-linearity allows
us to directly couple a background shear flow to the ac-
tive disturbance flow generated by a microswimmer. For
spherical microswimmers in a Poiseuille flow, we already
showed that due to such a coupling they experience a
swimming lift force that depends on the swimmer type
[17]. Here, we will show that this nonlinear coupling in-
fluences the Jeffery orbit of an elongated swimmer in a
shear flow, which thereby also fundamentally alters the
bulk rheological response. Such activity-induced changes
in the orientational dynamics cannot be observed in New-
 2

(a) (b) y tal stress tensor. It follows the second-order fluid (SOF)
model: T = −P I + 2 E + Wi S [34]. Here, E denotes the
V 1 = ˙ yex
<latexit sha1_base64="bhJ+f53R3xTLGr2yCQCcOIjdlp4=">AAACJHicbVDLSsQwFE19O76qLt0EB8HV0IqoIMKgG5cKzgOmdUgz6RhM0pLciqX0Y9z4K25c+MCFG7/FzGMxPg6EHM65N7n3RKngBjzv05manpmdm19YrCwtr6yuuesbTZNkmrIGTUSi2xExTHDFGsBBsHaqGZGRYK3o9mzgt+6YNjxRV5CnLJSkr3jMKQErdd3jIEpEz+TSXkWzvC4CrmLIS3yCg14CRdAnUpIS53iykJXd4r7sulWv5g2B/xJ/TKpojIuu+2bfpJlkCqggxnR8L4WwIBo4FaysBJlhKaG3pM86lioimQmL4ZIl3rFKD8eJtkcBHqqTHQWRZjCerZQEbsxvbyD+53UyiI/Cgqs0A6bo6KM4ExgSPEgM97hmFERuCaGa21kxvSGaULC5VmwI/u+V/5LmXs0/qO1f7lfrp+M4FtAW2ka7yEeHqI7O0QVqIIoe0BN6Qa/Oo/PsvDsfo9IpZ9yziX7A+foGpZ+mrg==</latexit>
<latexit sha1_base64="EMzZGwJtY7oVIh0YKPvwwmAZe6E=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIqMuiG5cV7APasWQymTY0kwxJRilD/8ONC0Xc+i/u/Bsz7Sy09UDI4Zx7yckJEs60cd1vp7Syura+Ud6sbG3v7O5V9w/aWqaK0BaRXKpugDXlTNCWYYbTbqIojgNOO8H4Jvc7j1RpJsW9mSTUj/FQsIgRbKz00A8kD/UktleWTAfVmlt3Z0DLxCtIDQo0B9WvfihJGlNhCMda9zw3MX6GlWGE02mln2qaYDLGQ9qzVOCYaj+bpZ6iE6uEKJLKHmHQTP29keFY59HsZIzNSC96ufif10tNdOVnTCSpoYLMH4pSjoxEeQUoZIoSwyeWYKKYzYrICCtMjC2qYkvwFr+8TNpnde+ifn53XmtcF3WU4QiO4RQ8uIQG3EITWkBAwTO8wpvz5Lw4787HfLTkFDuH8AfO5w9KEpMK</latexit>

p ∆
rate-of-strain tensor and S = 4E ⋅ E + 2δ E is the poly-
" meric stress tensor, which is quadratic in E and contains
Us !
<latexit sha1_base64="KjPx5yUrPVMO98KQD3iO9bcr4K0=">AAAB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkqMeiF48VTVtoQ9lsJ+3SzSbsboRS+hO8eFDEq7/Im//GbZuDtj4YeLw3w8y8MBVcG9f9dgpr6xubW8Xt0s7u3v5B+fCoqZNMMfRZIhLVDqlGwSX6hhuB7VQhjUOBrXB0O/NbT6g0T+SjGacYxHQgecQZNVZ68Hu6V664VXcOskq8nFQgR6NX/ur2E5bFKA0TVOuO56YmmFBlOBM4LXUzjSllIzrAjqWSxqiDyfzUKTmzSp9EibIlDZmrvycmNNZ6HIe2M6ZmqJe9mfif18lMdB1MuEwzg5ItFkWZICYhs79JnytkRowtoUxxeythQ6ooMzadkg3BW355lTQvqt5ltXZfq9Rv8jiKcAKncA4eXEEd7qABPjAYwDO8wpsjnBfn3flYtBacfOYY/sD5/AE9VI3I</latexit>

z the lower-convected time derivative. The SOF model

x not only allows us to capture the elastic effects perti-
nent to Boger fluids, but also to obtain analytical re-
sults for small Wi. Since we consider a steady shear rate
Fig. 1. Microswimmers in viscoelastic shear flow. in this work, we disregard the partial time derivative of
Schematics showing (a) the dilute suspension of microswim- E. In above governing equations, length, velocity, and
mers in an external shear flow V ∞ of a viscoelastic fluid, (b) pressure are already non-dimensionalized by the swim-
coordinate system moving with an individual swimmer that mer length (l = 2a), γ̇l, and µf γ̇, respectively, where µf
is modelled as an active prolate spheroid of major axis a and is the fluid shear viscosity. Also, the stress tensor is writ-
minor axis b. Here p and Us denote the orientation and swim- ten in dimensionless units using characteristic numbers
ming speed, respectively. (Wi and δ). In particular, Wi = trelax γ̇ is the shear based
Weissenberg number that quantifies the importance of
elasticity in the medium. It compares the shear rate γ̇
tonian fluids because the viscous stress tensor does not or inverse shearing time to the polymer relaxation time
permit such a nonlinear coupling. trelax = (Ψ1 + Ψ2 )/µf , where Ψ1 and Ψ2 are the normal
The current work uses the model of a second-order fluid stress coefficients. Furthermore, δ = −Ψ1 /2(Ψ1 + Ψ2 ) is
with a single elastic relaxation time that captures the the viscometric parameter that typically varies between
dynamics of polymeric fluids in the dilute limit (Boger −0.7 to −0.5. In ‘Methods: Hydrodynamic model’ we ex-
fluids) [34, 35]. For weak elasticity, quantified by the plain in detail how we solve the governing equations using
Weissenberg number, we perform a perturbative analy- a systematic perturbation expansion in Wi to address the
sis. Thereby we show that the inherent non-linearity in limit of weak viscoelasticity.
the polymeric stress tensor together with activity signifi- Active spheroid in viscoelastic shear flow. To
cantly alters the Jeffrey orbits known from the Newtonian determine the dynamics of an active spheroid, we first
fluid and of passive rods in a viscoelastic fluid [36, 37]. note that a swimmer disturbs the flow field both pas-
To evaluate the influence of thermal noise, we combine sively (due to its rigid body) and actively (due to self-
our results with the orientational Smoluchowski equation propulsion). The disturbance fields are implemented us-
and find that the altered deterministic dynamics also sub- ing hydrodynamic multipoles [39, 40]. Flagellated mi-
stantially affects the orientational distribution, known as croswimmers like E.coli and Chlamydomonas generate
the suspension microstructure [38]. It strongly differs a force-dipole flow field and higher order disturbances:
between extensile (pushers) and contractile (pullers) mi- source dipole, rotlet dipole, and force quadrupole. We
croswimmers. The orientational distribution allows us included all four of them and found that only the force-
to directly determine the effective viscosity of the active dipole flow field affects the swimmer dynamics in leading
suspension from an orientational average over the swim- order in Wi. The force-dipole field around the active
mer stresslets. Our analysis shows that fluid elasticity spheroid is σ rr̂2 [3(r̂ ⋅ p)2 − 1] with r̂ = r/r. Here, σ is
reduces the effective viscosity of active suspensions for a non-dimensional parameter equal to the ratio of the
both pushers and pullers compared to Newtonian fluids. force-dipole strength (σ ∗ ) to the stresslet imposed by the
The reduction increases with activity and allows to reach shear flow (8πµf γ̇l3 ), where σ > 0 represents pushers and
the superfluidic limit for smaller swimmer densities. σ < 0 pullers. For the current work, we focus on swim-
mers of 5µm size and shear rates of order 1−10s−1 , which
corresponds to σ of typical wild-type E.coli being roughly
RESULTS 0.04 − 0.1 [41, 42].
In Newtonian fluids, the equation of motion for the
Setup and second-order fluid. We consider a di- orientation p (θ, φ) gives the Jefferey orbits. To formu-
lute suspension of microswimmers in a shear flow of a late this equation for viscoelastic shear, we note that the
viscoelastic fluid, for instance, consisting of polymers polymeric stress tensor S is quadratic in the rate-of-strain
dissolved in a Newtonian fluid as shown in Fig. 1(a). tensor and vorticity [34]. Thus, similar to Ref. [43] we
Figure 1(b) depicts the coordinate system moving with determine all terms that by symmetry contribute to the
the microswimmer that is modelled as an active prolate rate of change ṗ up to first order in Wi. Neglecting small
spheroid and swims with speed Us . The uniformly dis- terms, we arrive at
tributed polymers are much smaller than the microswim-
mers and hence modeled within a continuum description. ṗ = (I − pp) ⋅ (E∞ ⋅ p) [Λ + Wi σα1 + Wi β1 E∞ ∶ pp]
The inertia-less hydrodynamics is governed by the mass + Ω∞ × p [1 + Wi σα2 + Wi β2 E∞ ∶ pp] + O(Wi2 ) (1)
and momentum conservation as ∇ ⋅ V = 0 and ∇ ⋅ T = 0,
respectively, where V is the velocity field and T is the to- in non-dimensional form, where Ω is the angular veloc-

(a) 𝑦 Log-rolling (b) 𝑦 Log-rolling (c) 𝑦
(Passive) (Weak Pusher)
𝑥 𝑥
𝑧 𝑧 𝑧
(d) 𝑦 Log-rolling (e) 𝑦 (f) 𝑦
Shear plane rotation
(Puller) (Strong Puller)
𝑥 𝑥
𝑧 𝑧 𝑧
Fig. 2. Orientation dynamics of passive (σ = 0) and active
particles in weakly viscoelastic shear flow. The blue curves
show time traces of the orientation vector p on the unit sphere
starting at the red dot. (a) σ = 0, (b) σ = 0.06, (c) σ = 0.2,
(d) σ = −0.2, (e) σ = −0.8, (f) σ = −2.4 . Other parameters:
Wi = 0.2, λ = 5, θ0 = π/5, φ0 = π/2, α1 = 4.62, α2 = −0.33,
β1 = 0.68, and β2 = 1.46, δ = −0.6.
ity and the superscript ∞ denotes the quantities that
belong to the prescribed shear flow. The shape factor
2
Λ = −1+λ contains the aspect ratio λ = a/b, the ratio
1+λ2
of major to minor axis. The shape factor approaches
+1 and −1 for needle and disk-like particles, respectively.
Since microswimmers are usually elongated, we focus on
prolate spheroids of Λ > 0.9, which corresponds to λ > 4;
it also helps in simplifying the calculations (see ‘Meth-
ods: Hydrodynamic model’). The terms with coefficients
αi and βi represent the active and passive viscoelastic
contributions, respectively. These coefficients are evalu-
ated explicitly as outlined in ‘Methods: Orientation dy-
namics’ using the Lorentz reciprocal theorem, where we
also derive Eq. (1). For our relevant parameters, we find
α1 ≫ ∣α2 ∣. Since the coefficients vary only weakly with
λ and δ, they are treated as constants in the following
discussion of results (see also Supplementary Note 2A).
In the absence of polymers (Wi = 0), Eq. (1) reduces
to the well-known Jeffery equation [25] that has an in-
finite number of neutrally stable solutions, i.e., the mi-
croswimmer’s orientation traces periodic orbits that de-
pend on initial conditions. During orbital time period
T = 2π(λ + λ−1 )/γ̇, they spent the majority of their time
(∝ λ) aligned near the flow-vorticity plane. When poly-
mers are present, the passive viscoelastic effect breaks
the degeneracy of Jeffery orbits and the microswimmer
slowly drifts towards an alignment with the vorticity axis,
known as the “log-rolling” state [37, 44, 45]. Figure 2a
shows the curve traced by the orientation of a spheroid
on a unit sphere while drifting towards the vorticity axis.
This dynamics can be deduced from Eq. (1) for passive
spheroids (σ = 0) when the terms with coefficients β1
and β2 are present. In Supplementary Note 2C we also
compare our results of passive spheroids with Brunn [37].
Now, we illustrate the influence of activity. For weak
pushers (0 < σ ≲ 0.1), the orbits towards log rolling look
3
Alignment
(Pusher) 0.01
0
- 0.01
𝑥 -2 -1 0
Align.
𝜙!"
Alignment Align.
SPR
(V. Strong Puller)
LR
𝑥
𝜎
Fig. 3. State diagram showing the different states of the
orientational dynamics for three Wi values. With increas-
ing dipole strength σ from left to right, we observe align-
ment (Align.), shear-plane rotation (SPR), log rolling (LR)
and again alignment. In the alignment state, the alignment
angle φeq is plotted versus σ as solid lines. The dots on the
horizontal axis indicate the critical values σc at which align-
ment (Align.) occurs either to the left of the dashed-dotted
lines (σ < 0) or to the right (σ > 0). The dashed lines sepa-
rate SPR and LR from each other at σ = −1.93, −0.49, −0.19
for increasing Wi. The inset shows the variation of Lyapunov
exponent with σ. Other parameters are chosen as in Fig. 2.
similar to the ones of passive spheroids, albeit with a
larger aspect ratio (see Fig. 2b) since they exhibit more
skewed trajectories. This can directly be inferred from
Eq. (1), where activity (σ) in the term with Wi σα1 mod-
ifies the shape factor. Since α1 > 0, a pusher (σ > 0) effec-
tively increases Λ, which corresponds to a higher aspect
ratio. The opposite occurs for weak pullers (σ < 0) as
they behave like a passive spheroid with reduced aspect
ratio. Here the trajectories are more circular as shown in
Fig. 2d.
As the dipole strength of a pusher increases beyond a
critical value σc , the orientation drifts to the shear plane
(θ = π/2) and aligns at an angle φeq with the flow di-
rection (see Fig. 2c). We quantify this transition in Fig.
3 and show that σc (dots close to 0) decreases with in-
creasing Wi while φeq increases with both σ and Wi.
This dynamical behaviour can again be discerned from
Eq. (1), which essentially balances the effect of rotational
(Ω∞ ) and elongational (E∞ ) flow on p. Now, activity
together with viscoelasticity allows to control this bal-
ance. In particular, for our case of α1 ≫ ∣α2 ∣, we expect
the elongational flow to dominate the dynamics for in-
creasing activity. Indeed, when the effective shape factor
Λ+Wi σα1 exceeds one at a critical value σc , the dynamics
of p transitions from the orbital to the alignment state as
favored by the elongational flow. In ‘Methods: Dynami-
cal analysis 1’, we show for a modified dynamical system
that the relevant eigenvalue of the dynamical matrix be-
comes real at σc , as expected for such a transition, and
the corresponding eigenvector gives the alignment angle
φeq . As σ increases, φeq grows from zero and approaches

π/4 for large σ, since in this limit, the elongational flow
(E∞ ) with its principle axis along φ = π/4 completely
dominates the dynamics of p.
For pullers, Fig. 2e shows that increasing the activity
induces a transition from log rolling towards rotation in
the shear plane. This orbit is also observed for passive
oblate particles in viscoelastic flows [37, 44, 45] and acts
as an attracting limit cycle. We have performed a stabil-
ity analysis for the shear-plane rotation in ‘Methods: Dy-
namical analysis 2’ and calculated the Lyapunov stability
exponent L. It determines the exponential time variation
of the disturbed limit cycle. In the inset of Fig. 3, we
show the Lyapunov exponent as a function of σ. For weak
and moderate pullers, L > 0 shows that shear-plane ro-
tation is an unstable limit cycle, where the system drifts
towards stable log rolling. As the dipole strength of the
puller becomes more negative, L turns negative mean-
ing that shear-plane rotation is stable. Upon analysing
the orbital dynamics in the shear-plane rotation, we find
that increasing ∣σ∣ gradually slows down the swimmer’s
rotation near the y-axis, along which the flow gradient is
applied. Eventually, a transition to permanent alignment
with φeq = π/2 occurs, which can be similarly analyzed as
the alignment transition of pushers. However, here it oc-
curs at the effective shape factor Λ + Wi σc α1 = −1. With
further increasing ∣σ∣ the active spheroid tilts against the
flow, in contrast to pushers, and asymptotes at 3π/4,
which is the direction of the second principal axis of the
elongational flow (E∞ ). This behavior is illustrated in
Fig. 2f and Fig. 3. Finally, Fig. 3 also shows that as Wi
increases, the regime of shear-plane rotation shrinks.
Impact of noise on orientational dynamics. The
deterministic orientational dynamics is disturbed by two
types of stochastic reorientations of the swimming direc-
tion, which we now address with the help of the Smolu-
chowski equation. First, a bacterium tumbles, which is
triggered when the rotation of one of its flagella reverses
so that it leaves the flagellar bundle [46–48]. For a wild
type E.coli, tumbling occurs roughly every 1s, where it
attains a new random orientation. Although tumbling
is biased in the forward direction with a mean tumbling
angle of roughly 68.5○ , we model it to be unbiased for
computational ease because it only affects our results
marginally as suggested by Nambiar et al. [49]. Second,
thermal rotational diffusion continuously reorients a mi-
croswimmer but its effect is small compared to tumbling.
We now evaluate the orientational distribution of an
ensemble of non-interacting microswimmers in steady
state, caused by the three mechanisms discussed so far:
deterministic motion in background shear flow, tum-
bling, and rotational diffusion. The steady-state proba-
bility distribution function ψ(p) for the orientation vec-
tor of non-interacting microswimmers is governed by the
Smoluchowski equation [50]
1 [55]. We quantify this in ‘Methods: Kinetic model 1’ by
Pef ∇p ⋅ (ṗψ) − τ Dr ∇2p ψ + (ψ − ) = 0, (2)
4π
4
where ψ(p) is normalized to one. The first term de-
scribes the orientational drift using ∇p as the gradient
operator on the unit sphere. To compare the strength of
flow-induced reorientation to the mean time τ between
two tumbling events, we introduce the flow Péclet num-
ber Pef = γ̇τ . Tumbling away from the swimming di-
rection p is handled by the third term and rotational
diffusion by the second term. For bacteria, the latter is
typically small compared to tumbling since τ ∼ 1s [46]
and Dr ≲ 0.1s−1 , as an estimate from the Stokes-Einstein
relation shows. In restricting ourselves to Eq. (2), we
assume a spatially uniform system valid when hydrody-
namic and steric interactions with bounding surfaces can
be neglected [29, 40]. In particular, this means that the
mean length of persistent swimming, Us τ , where Us is the
swimming speed, is much smaller than the spatial extent
of the system. Finally, we consider weak fluid elastic-
ity with relaxation time trelax ≪ τ, Dr−1 , so that the fluid
relaxes faster than the time scales given by stochastic
reorientations [51]. Thus, we do not need to take into
account any memory in the rotational noise.
In the following, we want to emulate a rheological ex-
periment and vary the shear rate γ̇ via Pef , while the
fluid and swimmer properties are kept constant. There-
fore, in Eq. (1) for the rotational drift velocity ṗ, we
rewrite Wi as Pef De, where the Deborah number De =
trelax /τ compares the fluid relaxation time to τ . Further-
more, the second relevant parameter, Wiσ, does not ex-
plicitly depend on γ̇ but rather quantifies the strength
of activity relative to fluid elasticity. To vary activ-
ity independent of other parameters, we replace Wiσ by
Pea De/8π. Here, Pea is the signed activity Peclet num-
ber that is positive for pushers and negative for pullers.
Using Us ∼ σ ∗ /(µf l2 ) for the swimming speed with the
force dipole moment σ ∗ [41], we can write it in the familiar
form Pea ∼ Us τ /l. Thus, its magnitude compares the per-
sistence length to the body length l. A wild-type E.coli
typically has Pea ≲ 5 [46, 52]. We numerically solve Eq.
(2) for arbitrary P ef by expanding Ψ in spherical har-
monics and taking into account the first 100 harmonics.
The numerical solution is also verified analytically in the
limit of Pef ≪ 1 (see further details in ‘Methods: Kinetic
model’).
Figure 4 shows the orientational probability distribu-
tion of passive and active particles for various cases. It
quantifies the orientational ‘microstructure’ of a suspen-
sion of non-interacting and orientable particles subject
to shear flow [49, 53, 54]. We begin with discussing the
case of weak shear rate (Pef < 1) in Figs. 4a-d. In the ab-
sence of shear flow, the microstructure is solely governed
by rotational noise and is therefore isotropic. For weak
shear rates the extensional part (E∞ ) of the shear flow
primarily distorts the microstructure, which peaks near
the extensional axis as noted for Newtonian fluids in Ref.
solving Eq. (2) via a perturbation expansion in Pef and
 5

obtain the first-order correction: 𝑦 𝑦

(a) (b) 0.128
3 E∞ ∶ pp Passive Pusher
ψ(1) = (Λ + α1 DePea /8π) (3) 0.113
4π 1 + 6τ Dr
0.098
In the absence of activity, there is no deviation from the 𝑥 𝑥
𝑧 𝑧 0.083
Newtonian microstructure at this order.
In Figs. 4a-d we use a weak shear rate of Pef = 0.5 0.068
and ∣Pea ∣ = 20, which is an activity close to mutated (c) 𝑦 (d)
strains of tumbling E. coli (pusher) [56] and the algae C. 0.14
Puller Pusher
reinhardtii (puller). The observed distributions follow 0.12 Passive

PDF
the trend of Eq. (3). While pushers (Pea > 0) enhance 0.10 Puller
the alignment along the extensional axis of the applied 0.08
𝑥 Newtonian
shear flow, pullers (Pea < 0) weaken it. We note that the 𝑧 0.06
0 𝜋 π
peak of the orientational distributions in the shear plane 4 𝜙
2
π
(plot d) exhibits a small deviation from φ = π/4. This
is due to the rotational flow (Ω∞ ), which contributes to 𝑦 𝑦
(e) 𝜙 (f) 0.45
ψ(p) in second order in Pef [54]. Passive Pusher
As the shear rate increases, the deterministic dynamics 𝑡 0.35
becomes more visible. We illustrate this in Figs. 4e-h for
0.25
Pef = 5. Compared to Fig. 4a, the peak in ψ(p) for pas-
sive particles shifts towards the flow axis and becomes 𝑥 𝑥
𝑧 𝑧 0.15
more anisotropic along the flow-vorticity plane, as the
0.05
deterministic velocity ṗ is smallest in this plane. Eventu-
(g) 𝑦 (h)
ally, ṗ becomes zero in the deterministic log-rolling state 𝜙 0.30
0.25
observed in viscoelastic fluids (Fig. 2a). However, this 1.2 0.2
0.20
Puller 𝑡 0.15
1.0 0.10
state is not completely reflected in the distribution be- 00 𝜃
π
𝜋π
0.8

PDF
2

cause the slow relaxation towards the log-rolling axis is Pusher

0.6
always interrupted by the dominating stochastic reorien- Passive
𝑥 0.4 Newtonian
tations [51, 57, 58]. Note that due to the fluid elasticity, 𝑧 0.2
Puller

the peak of ψ(p) is slightly closer to the flow axis com- 0 𝜋 π

pared to the Newtonian case as Fig. 4h shows. Further- 4 𝜙

2
π

more, the inset therein also illustrates a broader distri-

bution in the flow-vorticity plane (φ = 0). These findings (i) 𝑦 (j) 𝑦
𝜙 0.22
Pusher Puller
are consistent with earlier studies on passive fibre suspen- No Tumble
sions at moderate shear rates [51, 57]. Further increas-
No Tumble 𝑡 0.18

ing Pef spreads the distribution even more in the flow-
vorticity plane and, ultimately, for Pef ≳ 100 the peak
develops along the vorticity axis, which is entirely rem-
iniscent of log rolling. However, this is a regime which
we cannot strictly capture as our analysis requires De,
Wi < 1, which means Pef < 10.
Figure 4f demonstrates that the activity of pushers in
conjunction with the higher shear rate at Pef = 5 makes
the distribution more anisotropic and, according to Fig.
4h, strongly focused at an angle φ = 18○ . This is close to
the alignment angle φeq = 21○ of the deterministic align-
ment state with the parameters Wi = 0.5, σ = 0.16 in
Fig. 3. Conversely, pullers reduce the anisotropy in the
orientational distribution as shown in Fig. 4g. Similar to
the passive case, we can directly relate this observation
to the deterministic velocity ṗ with parameters Wi = 0.5,
σ = −0.16 that indicate again the log-rolling state; the
resulting orbit is similar to that depicted in Fig. 2d.
Compared to the passive case, the orbit is more circular,
which we already related to a reduced effective aspect ra-
tio. Therefore, the orientation is less constrained to the
flow-vorticity plane (see also the inset of Fig. 4g) and, as
a result, the distribution is broader and less anisotropic
𝑥 0.14
𝑥 0.10
𝑧
𝑧
0.06
Fig. 4. Orientational probability distribution ψ(p) for
different parameters: (a-d) weak shear flow at Pef = 0.5, (e-
j) strong shear flow at Pef = 5. (a,e) Pea = 0, (b,f) Pea =
20, (c,g) Pea = −20 and always with Dr τ = 0.1, (d,h) ψ(p)
in the shear plane; the inset in (h) shows ψ(p) in the flow-
vorticity plane. Non-tumbling particles: (i) Pea = 100, (j)
Pea = −100. The insets in (e,g,j) show φ(t) for the shear-plane
rotation state corresponding to the respective distributions.
Other parameters are chosen as in Fig. 2 with De = 0.1. To
locate these distributions in Fig. 3, we give the corresponding
Weissenberg and activity numbers: (a-d) Wi = 0.05, σ = ±1.6;
(e-h) Wi = 0.5, σ = ±0.16; (i,j) Wi = 0.5, σ = ±0.8.
[59]. Consequently, the particle is less susceptible to the
effect of stochastic reorientations and thus, its distribu-
tion has a peak closer to the flow axis. Hence, unlike

active Newtonian suspensions [24], the microstructure of
microswimmers in a viscoelastic fluid is clearly sensitive
to their hydrodynamic signature being either pushers or
pullers.
We also examined the orientational distributions of mi-
croswimmers that do not tumble and only experience ro-
tational thermal noise. Therefore, in this case we elimi-
nate the last term on the right-hand side of Eq. (2) and
redefine our parameters Pef , Pea , and De replacing τ by
Dr−1 . For passive particles and activity Pea ∼ O(10), the
microstructure is qualitatively similar to the distributions
in Figs. 4a-h. However, microswimmers that do not tum-
ble can have larger persistence lengths [60]. In particular,
one can genetically modify E. Coli such that tumbling
does not occur [41, 56, 61]. Therefore, in Figs. 3i and j
we show the orientational distributions for Pea = ±100.
Compared to Fig. 4f, increasing the activity of a pusher
moves the peak in the distribution function (at φ = 34○ )
even closer to the alignment angle (φeq = 36○ ) of the de-
terministic alignment state. For pullers we observe an
even more drastic change upon increasing the activity.
The peak in the orientational distribution shifts to the
quadrant where the compressional part of the shear flow
occurs (see Fig. 4j). Interestingly, this is not due to
the deterministic alignment of pullers as the parameters
(Wi = 0.5, σ = −0.8) belong to the shear-plane rotation
state (see Fig. 3). As already explained in the discus-
sion of Fig. 3, the rotational velocity ṗ slows down near
φ = π/2 when the effective shape factor Λ + Wi σc α1 be-
comes negative. Therefore, the active particle resembles
a passive oblate spheroid [62]. Indeed, the inset of Fig. 4j
shows how the puller orientation spends more time near
φ = π/2. This again shows that even in the presence of
significant noise the microstructure is determined by the
deterministic dynamics.
Shear viscosity of active suspensions. Finally, we
evaluate the effective shear viscosity of a dilute active
suspension from the total stress tensor, Σ = Σf + Σp ,
where subscripts f and p refer to fluid and particle con-
tributions, respectively. The deviatoric component of
Σ provides the shear viscosity, µ = µf + ϕµp , where
ϕµp = Σp xy /γ̇ is due to the suspended microswimmers
and ϕ = na3 is proportional to their volume fraction with
n being the particle density. Note that the shear vis-
cosity µf of the second-order fluid does not depend on
the shear rate, while µp is dependent on it. To evaluate
µp , we need to average over the stresslets Π(p) gener-
ated by the suspended particles. Using the orientational
distribution, we obtain
Σp = n⟨Π⟩ = n ∫ Π(p)ψ(p)dp, (4)
S viscosity is smaller than µf . Conversely, pullers (green
where Π(p) is the sum of three contribu-
tions [24], which give the following stress ten-
sors: ΣA
p = nσ ∗ [⟨pp⟩ − I/3] /8π due to activity,
T
Σp = 3nkB T [⟨pp⟩ − I/3] due to thermal reorientations,
and ΣF p due to the resistance of passive particles to
shear. For Newtonian fluids, the latter was first derived
6
by Einstein [63] for spherical particles, and then later
generalized to elongated particles by Hinch and Leal
[38, 53]. For the second-order fluid, we follow Férec
et al. [64] and approximate it using the Geisekus form,
∇
ΣF 3
p = −µf nl Aγ̇ ⟨pp⟩/2, where A = π/6 ln(2λ) is a shape
factor to access the friction of a long slender body
∇
and ⟨pp⟩ is the upper-convected time derivative of
the second moment of ψ(p). The expression for ΠF
was originally derived for dumbbells suspended in a
Newtonian fluid [65]. As in Ref. [64] we use it here as
an approximation for the stress response of particles in
a weakly viscoelastic fluid, as there are no closed form
expressions available. The consequences of the second-
order fluid come in through the orientational dynamics
ṗ in Eq. (1) and further terms in the upper-convected
derivative (see ‘Methods: Rheology’). The expressions
for the three contributions to the particle stress tensor
are pertinent to studies on slender rods and fibres (i.e.,
Λ → 1), and henceforth, we will be focusing on spheroids
with larger aspect ratios (λ = 10). Using Eq. (4) in the
definition of the particle-induced contribution to shear
viscosity, ϕµp = Σp xy /γ̇, together with our characteristic
parameters, we obtain
µp ∇ 8 Pea I
= −4A⟨pp⟩xy + (6τ Dr A − ) [⟨pp⟩ − ] .
µf Pef 8π 3 xy
(5)
We use numerical solutions of Eq. (2) for the orienta-
tional distribution ψ(p) to evaluate µp for varying Pef
and Pea . We also match the numerical values of µp with
an expression, which we obtain using the analytic form of
ψ(p) from Eq. (3) in the limit of small Pef (the derivation
is provided in ‘Methods: Rheology’). In Supplementary
Note 3, we also show that our results agree with Saintil-
lan [24] in the Newtonian limit.
Figure 5a shows the particle-induced contribution to
shear viscosity normalized by the bare fluid viscosity µf ,
which is plotted versus shear strength Pef . We begin
with discussing suspensions in a Newtonian fluid (dashed
lines) [24]. When a suspension of passive rods (black) is
sheared weakly (Pef ≪ 1), the orientational microstruc-
ture is governed by Eq. (3) with De = 0 and resembles
Fig. 4a. Here, the rods are aligned along the exten-
sional axis of the applied shear flow and resist shearing
(µp > 0), which enhances viscosity. For a suspension of
pushers (brown dashed) in the weak-shear regime, the
microstructure still resembles Fig. 4a, but now the ex-
tensile force dipoles support the elongational part of the
shear flow, which results in µp < 0 so that the total shear
dashed) with their contractile force dipoles additionally
resist shear flow and thereby enhance viscosity. Note
that, since the microstructure of active suspensions in a
Newtonian fluid is unchanged by the hydrodynamic sig-
nature of microswimmers, the magnitude of the activity
contribution to µp is identical for pushers and pullers.
As the shear rate increases, the magnitude of µp for both
 7

passive and active rods reduces, which resembles a typical (a) Pea = 20 (b) Pea = 20
<latexit sha1_base64="bTZz+WwKwB716ACCbRZGxBW+yj0=">AAACCnicbVBNS8NAEN34WetX1aOXaBG8WJJS1ItQ9OKxgv2ApoTNdtIu3WzC7kQsoWcv/hUvHhTx6i/w5r8xaXuorQ8GHu/NMDPPiwTXaFk/xtLyyuraem4jv7m1vbNb2Ntv6DBWDOosFKFqeVSD4BLqyFFAK1JAA09A0xvcZH7zAZTmobzHYQSdgPYk9zmjmEpu4chBeMSkBiOXXpUtx8nPCmdlK+8WilbJGsNcJPaUFMkUNbfw7XRDFgcgkQmqddu2IuwkVCFnAkZ5J9YQUTagPWinVNIAdCcZvzIyT1Kla/qhSkuiOVZnJxIaaD0MvLQzoNjX814m/ue1Y/QvOwmXUYwg2WSRHwsTQzPLxexyBQzFMCWUKZ7earI+VZRhml4Wgj3/8iJplEv2ealyVylWr6dx5MghOSanxCYXpEpuSY3UCSNP5IW8kXfj2Xg1PozPSeuSMZ05IH9gfP0CPWuZUQ==</latexit>
<latexit sha1_base64="Xw+5DnJqIYknP7knMTGDYnLT/qQ=">AAAB+HicbVBNS8NAEN3Ur1o/GvXoJVgETyUpRb0IRS8eK9gPaEPYbKft0s0HuxOxhv4SLx4U8epP8ea/cdPmoK0PBh7vzTAzz48FV2jb30ZhbX1jc6u4XdrZ3dsvmweHbRUlkkGLRSKSXZ8qEDyEFnIU0I0l0MAX0PEnN5nfeQCpeBTe4zQGN6CjkA85o6glzyz3ER4xbcLMo1c1u+SZFbtqz2GtEicnFZKj6Zlf/UHEkgBCZIIq1XPsGN2USuRMwKzUTxTElE3oCHqahjQA5abzw2fWqVYG1jCSukK05urviZQGSk0DX3cGFMdq2cvE/7xegsNLN+VhnCCEbLFomAgLIytLwRpwCQzFVBPKJNe3WmxMJWWos8pCcJZfXiXtWtU5r9bv6pXGdR5HkRyTE3JGHHJBGuSWNEmLMJKQZ/JK3own48V4Nz4WrQUjnzkif2B8/gDCopKB</latexit>

shear-thinning/thickening behaviour. For passive parti- Pea = 20

Newt. Puller
cles the microstructure is similar to Fig. 4e. The align- 0
Passive
ment close to the flow axis explains why the resistance of 𝜇! Flow
0.4 - 0.5
particles to the applied shear flow is reduced. For active 0.2 𝜇" Thermal

particles, the activity-induced flow becomes less and less- 0.20 Pea =-20 Active
Newt. Pusher
-1
important with increasing shear rate as quantified by the- 0.4 Peaa =20
Pe 20
<latexit sha1_base64="bTZz+WwKwB716ACCbRZGxBW+yj0=">AAACCnicbVBNS8NAEN34WetX1aOXaBG8WJJS1ItQ9OKxgv2ApoTNdtIu3WzC7kQsoWcv/hUvHhTx6i/w5r8xaXuorQ8GHu/NMDPPiwTXaFk/xtLyyuraem4jv7m1vbNb2Ntv6DBWDOosFKFqeVSD4BLqyFFAK1JAA09A0xvcZH7zAZTmobzHYQSdgPYk9zmjmEpu4chBeMSkBiOXXpUtx8nPCmdlK+8WilbJGsNcJPaUFMkUNbfw7XRDFgcgkQmqddu2IuwkVCFnAkZ5J9YQUTagPWinVNIAdCcZvzIyT1Kla/qhSkuiOVZnJxIaaD0MvLQzoNjX814m/ue1Y/QvOwmXUYwg2WSRHwsTQzPLxexyBQzFMCWUKZ7earI+VZRhml4Wgj3/8iJplEv2ealyVylWr6dx5MghOSanxCYXpEpuSY3UCSNP5IW8kXfj2Xg1PozPSeuSMZ05IH9gfP0CPWuZUQ==</latexit>

- 0.6 Pea = 20
inverse proportionality to Pef in Eq. (5). 0.1 1 10
0.1 0.5 1 5 10
Pe Pe ff
Pe
<latexit sha1_base64="9PH/Jr7+HnYtOooQHn0fOkV09xg=">AAAB8nicbVBNS8NAEN3Ur1q/qh69LBbBU0mkqMeiF48V7Ae0oWy2k3bpZhN2J2IJ/RlePCji1V/jzX/jts1BWx8MPN6bYWZekEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmTjVHJo8lrHuBMyAFAqaKFBCJ9HAokBCOxjfzvz2I2gjYvWAkwT8iA2VCAVnaKVuD+EJswZM+2G/XHGr7hx0lXg5qZAcjX75qzeIeRqBQi6ZMV3PTdDPmEbBJUxLvdRAwviYDaFrqWIRGD+bnzylZ1YZ0DDWthTSufp7ImORMZMosJ0Rw5FZ9mbif143xfDaz4RKUgTFF4vCVFKM6ex/OhAaOMqJJYxrYW+lfMQ042hTKtkQvOWXV0nroupdVmv3tUr9Jo+jSE7IKTknHrkidXJHGqRJOInJM3klbw46L86787FoLTj5zDH5A+fzB6rQkYQ=</latexit> <latexit sha1_base64="9PH/Jr7+HnYtOooQHn0fOkV09xg=">AAAB8nicbVBNS8NAEN3Ur1q/qh69LBbBU0mkqMeiF48V7Ae0oWy2k3bpZhN2J2IJ/RlePCji1V/jzX/jts1BWx8MPN6bYWZekEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmTjVHJo8lrHuBMyAFAqaKFBCJ9HAokBCOxjfzvz2I2gjYvWAkwT8iA2VCAVnaKVuD+EJswZM+2G/XHGr7hx0lXg5qZAcjX75qzeIeRqBQi6ZMV3PTdDPmEbBJUxLvdRAwviYDaFrqWIRGD+bnzylZ1YZ0DDWthTSufp7ImORMZMosJ0Rw5FZ9mbif143xfDaz4RKUgTFF4vCVFKM6ex/OhAaOMqJJYxrYW+lfMQ042hTKtkQvOWXV0nroupdVmv3tUr9Jo+jSE7IKTknHrkidXJHGqRJOInJM3klbw46L86787FoLTj5zDH5A+fzB6rQkYQ=</latexit>

Now, we discuss the results for the second-order fluid Pef |Pea | = 20 f
<latexit sha1_base64="iZiCXOrFhYK1E3r07PDdat8RYEQ=">AAACDnicbVDJSgNBEO2JW4xb1KOXxhDwFGYkqBch6MVjBLNAJoSeTk3S2LPQXSOGSb7Ai7/ixYMiXj1782/sLKAmPih4vFdFVT0vlkKjbX9ZmaXlldW17HpuY3Nreye/u1fXUaI41HgkI9X0mAYpQqihQAnNWAELPAkN7/Zy7DfuQGkRhTc4iKEdsF4ofMEZGqmTLw5dhHtMqzDqsOH5sU1dN/cj+dSVkjqdfMEu2RPQReLMSIHMUO3kP91uxJMAQuSSad1y7BjbKVMouIRRzk00xIzfsh60DA1ZALqdTt4Z0aJRutSPlKkQ6UT9PZGyQOtB4JnOgGFfz3tj8T+vlaB/1k5FGCcIIZ8u8hNJMaLjbGhXKOAoB4YwroS5lfI+U4yjSTBnQnDmX14k9eOSc1IqX5cLlYtZHFlyQA7JEXHIKamQK1IlNcLJA3kiL+TVerSerTfrfdqasWYz++QPrI9v8oObZQ==</latexit>

(SOF) in Fig. 5a (solid lines). For the suspension of (c) No Tumble (d) No Tumble: |Pea | = 100
<latexit sha1_base64="6eDU1rJeW3aHk2ve6FuP/JMes/g=">AAAB+nicbVBNS8NAEN3Ur1q/Uj16CRbBU0lE1ItQ9OKxgv2ANoTNdtIu3XywO1FL2p/ixYMiXv0l3vw3btsctPXBwOO9GWbm+YngCm372yisrK6tbxQ3S1vbO7t7Znm/qeJUMmiwWMSy7VMFgkfQQI4C2okEGvoCWv7wZuq3HkAqHkf3OErADWk/4gFnFLXkmeVxF+EJszpMPDq+cmzbMyt21Z7BWiZOTiokR90zv7q9mKUhRMgEVarj2Am6GZXImYBJqZsqSCgb0j50NI1oCMrNZqdPrGOt9KwglroitGbq74mMhkqNQl93hhQHatGbiv95nRSDSzfjUZIiRGy+KEiFhbE1zcHqcQkMxUgTyiTXt1psQCVlqNMq6RCcxZeXSfO06pxXz+7OKrXrPI4iOSRH5IQ45ILUyC2pkwZh5JE8k1fyZoyNF+Pd+Ji3Fox85oD8gfH5A8swk7I=</latexit>

Pef ⌧ 1
passive rods (black), we do not find an appreciable de- Newt.
VE Puller
viation from the Newtonian case until Pef ≈ 10, similar Newt. Puller
VE Pusher
𝜇! Newt. Pusher
to Refs. [51, 57]. However, for active suspensions the 0.4
𝜇" 0.2
modifications are substantial and qualitatively different Newt.
0 Pea =-20
VE Puller
for pushers (brown) and pullers (green). For pushers, VE Pusher
- 0.2 Puller
- 0.4 Pusher
Pea =20
the reduction in viscosity is further enhanced compared - 0.6
to Newtonian fluids, whereas the viscosity enhancement Pea 0.1 0.5 1 5 Pe
10
f
<latexit sha1_base64="TrZq5lDQotV2Bm76pCEL/gElgXQ=">AAAB8nicbVBNS8NAEN3Ur1q/qh69LBbBU0mkqMeiF48V7Ae0oWy2k3bpZhN2J2IJ/RlePCji1V/jzX/jts1BWx8MPN6bYWZekEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmTjVHJo8lrHuBMyAFAqaKFBCJ9HAokBCOxjfzvz2I2gjYvWAkwT8iA2VCAVnaKVuD+EJswZM+6xfrrhVdw66SrycVEiORr/81RvEPI1AIZfMmK7nJuhnTKPgEqalXmogYXzMhtC1VLEIjJ/NT57SM6sMaBhrWwrpXP09kbHImEkU2M6I4cgsezPxP6+bYnjtZ0IlKYLii0VhKinGdPY/HQgNHOXEEsa1sLdSPmKacbQplWwI3vLLq6R1UfUuq7X7WqV+k8dRJCfklJwTj1yROrkjDdIknMTkmbySNwedF+fd+Vi0Fpx85pj8gfP5A6M8kX8=</latexit>

<latexit sha1_base64="9PH/Jr7+HnYtOooQHn0fOkV09xg=">AAAB8nicbVBNS8NAEN3Ur1q/qh69LBbBU0mkqMeiF48V7Ae0oWy2k3bpZhN2J2IJ/RlePCji1V/jzX/jts1BWx8MPN6bYWZekEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmTjVHJo8lrHuBMyAFAqaKFBCJ9HAokBCOxjfzvz2I2gjYvWAkwT8iA2VCAVnaKVuD+EJswZM+2G/XHGr7hx0lXg5qZAcjX75qzeIeRqBQi6ZMV3PTdDPmEbBJUxLvdRAwviYDaFrqWIRGD+bnzylZ1YZ0DDWthTSufp7ImORMZMosJ0Rw5FZ9mbif143xfDaz4RKUgTFF4vCVFKM6ex/OhAaOMqJJYxrYW+lfMQ042hTKtkQvOWXV0nroupdVmv3tUr9Jo+jSE7IKTknHrkidXJHGqRJOInJM3klbw46L86787FoLTj5zDH5A+fzB6rQkYQ=</latexit>

of pullers is reduced. We understand this using the ori- Pef

entational distributions in Fig. 4. Pushers are even more
aligned in the extensional quadrant of the applied shear
flow (Fig. 4b,d,f), which explains their further increased
support of shear flow. In contrast, pullers are less aligned
compared to the Newtonian case (Fig. 4c,d,g) and, there-
fore, oppose shear flow less. Figure 5b for pushers clearly
A limit Pef → 0 for tumbling and non-tumbling active rods.
shows that the active stress (Σ ) dominates and hence is
primarily responsible for the observed viscous response.
In the limit of shear rate much smaller than tumbling
rate, Pef ≪1 , we can calculate µp /µf analytically using
Eq. (3) (see ‘Methods: Rheology’) and obtain:
Fig. 5. Particle-induced contribution to viscosity
µp /µf in a second-order fluid. (a) Plotted vs. shear rate
Pef for pusher (Pea > 0), puller (Pea < 0), and a passive
rod (Pea = 0) with solid lines. Dashed lines correspond to
the Newtonian case. (b) Contributions to particle-induced
viscosity for Pea = 20. (c) µp /µf vs. activity Pea in the
(d) µp /µf vs. shear rate Pef for non-tumbling active rods
with large persistence length, ∣Pea ∣ = 100. Dotted black lines
near Pef = 0.1 in (a,d) correspond to the analytical results for
Pef ≪ 1. Other parameters: De = 0.1, λ = 10, α1 = 5.92, α2 =
−0.91 β1 = 0.63, β2 = 1.18.

µp 4A(5 − 3Λ) 48Aτ Dr − Pea /π De Pea α1

= + (Λ + ) 5c, dashed lines). Non-tumbling microswimmers can ex-
µf 15 5 (1 + 6τ Dr ) 8π
De Pea Aα1
− . (6)
10π
We discuss the result here since the influence of passive
and active rods on the shear viscosity is the strongest in
this limit. In Fig. 5c we plot µp /µf vs. Pea for tumbling
microswimmers (solid lines). As Eq. (6) shows, elasticity
in the fluid adds a quadratic dependence in Pea , while in
a Newtonian fluid the dependence is only linear [24]. For
all Pea , except a small region close to Pea = 0, the values
of µp /µf remain below those obtained in the Newtonian
limit. Thus, fluid elasticity reduces the shear viscosity
for both pusher and puller suspensions. Note that, as
we increase the activity of pullers (Pea ≲ −50), they too
can reduce the shear viscosity (µp /µf < 0) like pushers.
This is a consequence of the orientational distribution
shown in Fig. 4j; pullers with large activity align prefer-
entially in the compressional quadrant of the shear flow
and thereby also support the shearing fluid similar to
pushers aligned in the extensional quadrant.
As described earlier, we can treat the case of non-
tumbling microswimmers by replacing τ by Dr−1 in the
definitions of Pef , Pea , and De. With these parame-
ters, Eq. (6) can be formulated in the limit τ → ∞ to
show that the viscosity contribution of non-tumbling ac-
tive rods also follows a quadratic dependence in Pea (Fig.
hibit high persistence lengths. Thus, in Fig. 5d we show
for ∣Pea ∣ = 100 that their contribution to viscosity is neg-
ative over a wide range of shear rates for both pushers
and pullers in a second-order fluid. Hence, we find that
the elasticity of a fluid always results in a reduced to-
tal viscosity as compared to suspensions in a Newtonian
fluid.
DISCUSSION
In this article we show how activity influences the dy-
namics of a sheared microswimmer suspension in a vis-
coelastic fluid at an individual level and in the bulk. At
the individual level, the orientational dynamics of pas-
sive rods is well known from Jeffery’s orbits in Newtonian
shear [25] and ‘log-rolling’ orbits in viscoelastic shear flow
[36, 37]. Our analytical result [Eq. (1)], derived for elon-
gated active spheroids in a second-order fluid, demon-
strates how the active flow field of a swimmer modifies the
orientational dynamics. Extensile swimmers like E.coli
(pushers) drift to the shear plane and align at an angle
to the flow direction. For contractile swimmers like C.
reinhardtii (pullers), activity effectively lowers their as-
pect ratio. With increasing dipole strength, pullers show
log rolling, transition to shear-plane rotation, and ulti-

mately align at an angle against the flow direction. The
latter occurs for strong pullers at dipole strengths rele-
vant to artificial swimmers with large propulsion speed
[66, 67].
To demonstrate the impact of the individual swim-
mer dynamics on the bulk rheological behaviour, we em-
ploy the Smoluchowski equation to evaluate the orien-
tational probability distribution of an ensemble of ac-
tive spheroids called the suspension microstructure [38].
Accounting for tumbling and rotary diffusion, we find
that the microstructure substantially depends on the
hydrodynamic signature of the swimmer unlike suspen-
sions in Newtonian fluids [22, 24, 49]. Pushers align
more strongly in the extensional quadrant of the ap-
plied shear flow, while the alignment of pullers is sig-
nificantly weaker. This activity-specific microstructure
significantly modifies the effective shear viscosity com-
pared to Newtonian fluids [21]. In particular, the viscos-
ity reduction of a dilute pusher suspension is more pro-
nounced under viscoelastic shear flow, while the viscosity
enhancement of pullers is weaker. Thus, the activity of
a microswimmer contributes in two ways to the rheology
of swimmer suspensions in a viscoelastic fluid; directly
through its active stresses and indirectly by coupling to
the elasticity of the fluid and thereby influencing the ori-
entational microstructure. In particular, in the weak-
shear limit the particle-induced contribution to viscosity
scales quadratically with the activity Pea . In total, we
find that the elasticity of a second-order fluid always re-
duces the total viscosity of the microswimmer suspension.
Especially for pushers, this can help to reach the regime
of superfluidity at lower volume fractions compared to
Newtonian fluids.
Biologically relevant fluids are viscoelastic. In this ar-
ticle we presented a first systematic study of the individ-
ual dynamics and the bulk rheology of microswimmers
suspended in such fluids. Earlier investigations on ac-
tive suspensions in quiescent [68] and vortical viscoelastic
flows [69] assumed conventional Newtonian Jeffery orbits.
In the light of our results, these studies on the collec-
tive behaviour need to be revisited. Our investigations
makes several assumptions to address the complexity of
microbial flows, which offers ample opportunities for fu-
ture developments. Biological fluids can be more complex
and exhibit shear-thinning/-thickening properties or sev-
eral characteristic relaxation times, which requires more
evolved modeling using, for example, the FENE-P or
Giesekus model. Furthermore, mucus besides being sig-
nificantly shear-thinning has relaxation times larger than
1 second, which is of the order of the bacterial mean
free tumble time [15, 28, 29]. Thus, noise becomes non-
Markovian and memory needs to be incorporated in the
stochastic description, for example, within a generalized
Langevin equation [70].
8
METHODS
Hydrodynamic model
The hydrodynamics around the spheroidal particle is
governed by the mass and momentum conservation as
∇ ⋅ V = 0, ∇ ⋅ T = 0. Here T is the total stress tensor de-
∆
fined as T = −P I + 2 E + Wi S [34], where S = 4E ⋅ E + 2δ E.
The lower-convected derivative is defined as
∆ ∂E
E=[ + (V ⋅ ∇) E + E ⋅ ∇V + ∇V T ⋅ E] . (7)
∂t
Since we consider a steady shear flow and a steadily active
swimmer, we can disregard the temporal derivative in the
above stress tensor. In particle frame of reference, the
boundary condition on its surface is the no-slip condition
V = Ωp × r, where the rotational velocity Ωp is currently
unknown and will be evaluated by using the torque-free
condition. We split the velocity field into the disturbance
field v and background flow field V ∞ = E∞ ⋅r+Ω∞ ×r i.e.,
V = v + V ∞ . The equations governing the disturbance
flow field are
∇ ⋅ v = 0, −∇p + ∇2 v = −Wi (∇ ⋅ s), where
∆ ∆
s = 4(e ⋅ e + w) + 2 δ( e + w). (8)
Here e is the rate of strain tensor for the disturbance
∆ ∆
flow (∇v + ∇v † )/2, whereas w, e and w are the different
constituents of the polymeric tensor (s) due to the distur-
bance flow field. Specifically, w is the interaction tensor
defined as E∞ ⋅ e + e ⋅ E∞ , arising from the interaction
∆
between background flow and disturbance field; and w is
the lower-convected derivative of e and E∞ with respect
to V ∞ and v, respectively:
∆
w = V ∞ ⋅ ∇e + e ⋅ ∇V ∞ † + ∇V ∞ ⋅ e (9)
+ v ⋅ ∇E∞ + E∞ ⋅ ∇v † + ∇v ⋅ E∞ .
The boundary conditions of the disturbance field at par-
ticle surface and far away are
v = ∆Ω × r − E∞ ⋅ r at r ∈ S and v → 0 as r → ∞,
(10)
respectively. Here ∆Ω is the difference in particle angular
velocity and background vorticity (Ωp − Ω∞ ).
For small values of Wi, the disturbance field variables
are expanded as f = f (0) +Wi f (1) +⋯. Here, f is a generic
field variable that represents velocity (v), pressure (p),
and angular velocity (Ωp ). We substitute this expansion
in the Eq. (8) and obtain the O(1) Stokes problem as
∇ ⋅ v (0) = 0, −∇p(0) + ∇2 v (0) = 0 (11)
with the boundary condition at particle surface:
v (0) = ∆Ω(0) × r − E∞ ⋅ r at r ∈ S, (12)
and a decaying condition at infinity (v (0) → 0). Using
the finite multipole expansion around the spheroid [39,
 9

43, 71], we obtain the disturbance velocity for a passive with a decaying condition at infinity (v (1) → 0). Conven-
spheroid at O(1) as tionally, a solution to Eqs. (17) and (18) (i.e., v (1) ) is
(0)
sought, which, on the implementation of torque-free con-
vi = QR R R
ij,k εjkl {[A pl pm + B (δlm − pl pm )] ∆Ωm (1)
dition, reveals the modification of Jeffery orbits (Ωp ).
+ CR εlmn pm E∞
no po }
We employ the reciprocal theorem [71, 72] to avoid solv-
ing for the first order flow field and directly obtain the
∞ rotation velocity as
ij,k + χQij,llk ) {[A pjklm + B pjklm + C pjklm ] Elm
+ (QS Q S A S B S C

− CR (εjlm pk pm + εklm pj pm ) ∆Ωl }. (13) t

p ⋅T =∫
Ω(1) s(0) ∶ ∇v t dV. (19)
Vf

Here Q represent the spheroidal multipoles i.e. integral
representation of fundamental singularities spread on a
line √extending from one foci to another (−c to c, where
c = λ2 − 1/λ); QR , QS represent rotlet and stresslet
around the spheroid, respectively. Here χ = 8(λ12 −1) rep-
resents the strength of higher order quadrupolar field
(QQij,llk ). Since we focus on elongated particles, we ex-
clude this component for computational ease (λ = {5, 10}
correspond to χ = {0.005, 0.001}). Following Einarsson
et al. [43], we simplify these multipoles as
Here, superscript t refers to the test/auxiliary problem
where the particle rotates at a unit velocity Ωt = ei ,
where i corresponds to either of the three Cartesian co-
ordinates. The test torque is:
64πc3 R
T t = R ⋅ Ωt , where R = [A pp + BR (I − pp)] (20)
3
We solve Eq. (19) for three test field torques (along three
Cartesian coordinates) and obtain the following system
of equations: R† ⋅ Ωp = {I1 , I2 , I3 }, where † represents
(1)

transpose and Ii is the solution of volume integral (19)

QR 0 1
ij,k = (δik rj − δij rk ) J3 + (δij pk − δik pj ) J3 (14a)
for the test field in ith direction. We further simplify the
QSij,k = δjk ri J30 − δjk pi J31
− 3ri rj rk J50 (14b) above relation by noting that R is a symmetric matrix
(1)
1
+ 3(pi rj rk + pj ri rk + pk ri rj )J5 and ṗ(1) = Ωp × p:
− 3(ri pj pk + rj pi pk + rk pi pj )J52 + 3pi pj pk J53 . ṗ(1) = (R−1 ⋅ {I1 , I2 , I3 }) × p (21)
Here I and J represent various integrals defined as Eq. (21) is evaluated over a discretized angular grid,
where each point requires solving the volume integral in
c ξb Eq. (19). The polymeric stress therein (s(0) ) is given by
Iab = ∫ dξ; Jab = c2 Iab − Iab+2 . (15)
−c ∣r − ξp∣a Eq. (8) and is entirely dependent on O(1) disturbance
flow fields i.e., passive disturbance in Eq. (13) and active
In Eq. (13), the fourth order orientation tensors (pA , pB , disturbance in Eq. (16). We find that out of all active
pC ) and other coefficients (A, B, C) are described in the fields, only force-dipole (∼ r−2 ) provides a modification
Supplementary Note 1. at the current order of approximation; the rest decays in
At O(1), we add the activity using the far-field descrip- odd powers of distance (∼ r−3 ), and due to antisymmetry,
tions, which consists of a force-dipole (FD), source-dipole give zero contribution to the volume integral at O(W i).
(SD), rotlet-dipole (RD), and force-quadrupole (FQ) [40]:
Orientation dynamics
(0) −r 3r (r ⋅ p)2 We evaluate ṗ(1) analytically by noting that it stems
vF D = σF D ( 3 + ) (16a)
r r5 from the leading order polymeric stress [specifically s(0)
in Eq. (19)]; its form in Eq. (8) suggests that the modifi-
(0) 3rr p
vSD = σSD ( − I) 3 (16b) cation of a passive particle will be quadratic in the flow
r2 2r
gradient tensor, which can be decomposed in symmetric
(0) 3p ⋅ r
vRD = σRD [ 5 (p × r)] (16c) E∞ and antisymmetric O∞ (rotation-rate tensor) compo-
r
nents. Following Einarsson et al. [43], this can be written
(0) p 3 5(p ⋅ r)3 r in the general form as:
vF Q = σF Q { 3
− 5 [3(p ⋅ r)r + (p ⋅ r)2 p − ]}
r r r2
(1,P ) (1) (2) (3)
(16d) ṗi jk Elm + Kijklm Ejk Olm + Kijklm Ojk Olm ,
= Kijklm E∞ ∞ ∞ ∞ ∞ ∞

(22)
At O(Wi), the governing equation is where superscript P denotes passive contribution. The
coefficients of the fifth-order tensor K are composed of
∇ ⋅ v (1) = 0, −∇p(1) + ∇2 v (1) = − (∇ ⋅ s(0) ), where all possible permutations of the orientation vector p with
∆(0) ∆ (0) E∞ and O∞ :
s(0) = 4(e(0) ⋅ e(0) + w(0) ) + 2 δ( e +w ), (17)
(i)
Kijklm = ∑ (βn[1] pn1 pn2 pn3 pn4 pn5 + βn[2] pn1 pn2 pn3 δn4 n5
The boundary condition at the particle surface being n=5!

v (1) = Up(1) + Ω(1) +βn[3] pn1 δn2 n3 δn4 n5 ) .

p ×r at r ∈ S, (18)

Here β represents the unique coefficients for all the 5!
terms.
When activity is added in the hydrodynamics, the form
of polymeric stress tensor in Eq. (8) suggests that the
gradients of disturbance velocity (directed with p) will
multiply with gradients of passive disturbance (which
originate from E∞ and O∞ ) to yield further modification
to ṗ(1) . Thus, the contribution resulting from interaction
of active disturbances (p) with background flow (E∞ and
O∞ ) takes the following form at O(Wi):
(1,A) (1) (2)
ṗi = σ [Lijk E∞
jk + Lijk Ojk ] ,
∞
where coefficients of the third-order tensor L are com-
posed of the following combinations
(i)
Lijk = ∑ (αn[1] pn1 pn2 pn3 + αn[2] pn1 δn2 n3 ) .
n=5!
Here α represent the unique coefficients for all the 3!
terms. We simplify Eq. (22) and (23) by using the sym-
metry arguments and noting that magnitude of p is al-
ways unity. We obtain the following six irreducible terms:
ṗ(1) = E∞ ∶ pp [β1 (I − pp) ⋅ (E∞ ⋅ p) + β2 O∞ ⋅ p]
+ (I − pp) ⋅ [β3 (E∞ ⋅ E∞ ) ⋅ p + β4 (O∞ ⋅ E∞ ) ⋅ p]
+ σα1 (I − pp) ⋅ (E∞ ⋅ p) + σα2 O∞ ⋅ p.
We calculate the coefficients (αi and βi ) by calculating
ṗ(1) numerically for six independent orientations (p) and
solve the system of equations from Eq. (24) to extract
the coefficients. We find that β3 and β4 are nearly zero
(≲ 10−5 ) and thus neglect them. We obtain the analytical
approximation as
ṗ(1) = (I − pp) ⋅ (E∞ ⋅ p) [σα1 + β1 E∞ ∶ pp]
+ O∞ ⋅ p [σα2 + β2 E∞ ∶ pp] ,
We use the above equation in the main text as Eq. (1),
where we write O∞ ⋅p as Ω∞ ×p. In Supplementary Note
2A, we show the weak variation of these coefficients with
particle aspect ratio λ and the viscometric coefficient δ.
Dynamical analysis
1. Eigenvalue analysis of alignment dynamics. Following
Bretherton [73], we explain the alignment of a particle
in the shear plane by extracting a fundamental matrix
solution of Eq. (1). For this, we first consider the non-
dimensional Jeffery equation in the expanded form as
ṗ = O∞ ⋅ p + Λ [E∞ ⋅ p − p(E∞ ∶ pp)] ,
which can be represented in an unconserved form that
facilitates a fundamental matrix solution
q̇ = (ΛE∞ + O∞ ) ⋅ q.
q(t)
Here ∣q(t)∣ = p(t) and ∣q(t)∣ = exp[Λ (E∞ ∶ pp) t] repre-
sents the exponential elongation of q. We study the an-
gular dynamics of q, as its normalization yields back the
orientation vector p. The solution to Eq. (27) is
q(t) = exp [(ΛE∞ + O∞ )t]q(0).
10
The eigensystem of the above exponential matrix deter-
mines the orbital dynamics of spheroid. For the two-
dimensional shear flow, the eigenvalues√ of the matrix
in the exponent (ΛE∞ + O∞ ) are: 0, ± Λ2 − 1/2. The
non-zero eigenvalues are purely imaginary as 0 < Λ < 1,
where Λ = 1 for an infinitely slender particle. As a con-
sequence of this imaginary pair, we observe degenerate
infinite solutions of orbits in Newtonian fluids. For the
case of pure elongational flow (O∞ ≡ 0), the eigenvalues
are real: 0, ±Λ/2, which reveals the absence of periodic
orbits. The normalized eigenvector corresponding to the
(23) positive eigenvalue
√ √ determines the equilibrium orienta-
tion: {1/ 2, 1/ 2, 0} i.e., 45° or 225° in the two exten-
sion quadrants.
We now use Eq. (1) for the second-order fluid and
consider only the active viscoelastic component because
the passive effects do not contribute to the alignment
[this assumption is later verified by matching the results
with numerical integration of complete Eq. (1)]. In this
case, the equation of motion for the unconserved vector
q is
q̇ = [E∞ (Λ + Wiσα1 ) + O∞ (1 + Wiσα2 )] ⋅ q, (29)
which yields the solution as
(24)
q(t) = exp [ [E∞ (Λ + Wiσα1 ) + O∞ (1 + Wiσα2 )] t]q(0).
(30)
We obtain the eigenvalues of the fundamental matrix as
√
1
0, ± Λ2 − 1 + 2σWi (α1 Λ − α2 ) + Wi2 σ 2 (α12 − α22 ).
2
1−Λ
These eigenvalues turn real when σ > Wi(α1 −α2 )
or σ <
(25) −1−Λ
Wi(α1 +α2 )
. For instance, for a particle of aspect ra-
tio λ = 5, the negative and positive critical values are
σcrit = −0.448/Wi; 0.016/Wi, which matches with Fig. 3
generated from the integration of full Eq. (1) (see Sup-
plementary Note 2B). In the alignment regimes of Fig.
3, we further find that the normalized eigenvector corre-
sponding to the positive eigenvalue matches the angle of
alignment (φeq ) obtained numerically.
In addition to validating our results of Fig. 3, Eq. (30)
provides a key insight that the contribution from active
disturbances is to effectively alter the strength of elon-
gation (prefactor of E∞ ) and the rotation rate (prefac-
(26) tor of O∞ ). The first prefactor is the more decisive as
α1 ≫ ∣α2 ∣. It suggests that a pusher (σ > 0) disturbs the
local shear flow in order to increase the weight of elon-
gation. Once the activity exceeds the critical limit, the
local shear flow transforms into an effective elongation
(27)
flow whose axis of elongation points in the direction of
the normalized eigenvector associated with the positive
eigenvalue. As pusher’s activity further increases, the lo-
cally elongated flow asymptotically approaches the pure
elongation state where the impact of O∞ is negligible in
comparison (similar to our discussion of O∞ ≡ 0 case in
(28) previous paragraph).
 11

2. Stability analysis of orbits. Here we find the stability Following [49, 54], we solve the above inhomogeneous
exponent of the complete non-linear Eq. (1) to analyze linear differential equation using the Green’s function
the onset of shear-plane rolling state as shown in Fig. G(p∣p′ ), which is governed by
2,3. For Newtonian fluids, θ = π/2 is one of the infinite
neutral Jeffery orbits. For SOF, the polymeric stress in − τ Dr ∇2p G(p∣p′ ) + G(p∣p′ ) = δ(p − p′ ), (37)
the fluid lifts this degeneracy. To quantify this, we write
down the angular dynamics as where δ represents the Dirac-delta function. Following
[75], we expand this G into spherical harmonic series as
θ̇ = f (θ, φ), φ̇ = g(θ, φ), (30a,b)
∞ n
and perform a Taylor expansion near the shear-plane: G(p∣p′ ) = ∑ ∑ Cn,m (p′ )Ynm (p). (38)
n=0 m=−n
θ(t) = π/2+(t). Here  represents the deviation from Jef-
fery orbit and and we determine its growth i.e., whether
Here Cn,m are the series coefficients and Ynm (p) repre-
it grows or shrinks with time. At O() we obtain
sent the spherical harmonics. Similarly, the Dirac-delta
d ∂f function can also be related to the spherical harmonics as
= ∣ , (31) δ(p − p′ ) = ∑∞ n m m m
n=0 ∑m=−n Yn (p)Yn (p ), where Yn repre-
′
dt ∂θ θ=π/2
sents the corresponding complex conjugate spherical har-
which can be simplified and integrated over an orbit to monic [75]. Eq. (37) is now expressible as
obtain
∞ n
2 m m
∑ ∑ Cn,m (p ){−τ Dr ∇p [Yn (p)] + Yn (p)} =
⎡ −2π ⎤ ′
⎢ 1 ∂f ⎥
 = 0 exp ⎢∫
⎢ ( ) dφ⎥⎥ . (32) n=0 m=−n
⎢ 0 g(θ, φ) ∂θ θ=π/2 ⎥ ∞ n
⎣ ⎦ m
∑ ∑ Yn (p)Ynm (p′ ). (39)
Here the integration is over −2π because the particle’s n=0 m=−n
rotation due to shear is in −φ direction. Eq. (32)
can be expressed in the form of Lyapunov exponent Using the properties of spherical harmonics [75], we sub-
( = 0 exp[LT ]) [74] as stitute ∇2p Ynm = −n(n+1)Ynm , take the inner product with
respect to YNM (p) on both sides of Eq. (39), and use the
−2π
1 1 ∂ θ̇ orthogonality property to obtain
L∼
T ∫0 [ ]
φ̇ ∂θ θ=π/2
dφ, (33)
Ynm (p′ )
Cn,m (p′ ) = , (40)
where T is the time period of a Jeffery orbit 2π(λ+λ−1 )/γ̇. 1 + τ Dr n(n + 1)
We show the solution to Eq. (33) in the inset of Fig. 3. √
where the normalization condition yields C0,0 = 1/ 4π.
Kinetic model
We finally use the Green’s function solution to find ψ(1) :
1. Near-equilibrium microstructure. First, we detail
the evaluation of microstructure near equilibrium (Pef =
3 E∞ ∶ pp
γ̇τ ≪ 1) i.e., when stochastic reorientation dominates the ψ(1) = (Λ + α1 DePea /8π) [ ]. (41)
shear-induced reorientation. In this limit, we expand the 4π 1 + 6τ Dr
orientation distribution as
In the limit of weak shear, this solution matches with the
ψ = ψ(0) + Pef ψ(1) + ⋯. (34) numerically obtained ψ for arbitrary Pef , whose evalua-
tion is discussed next.
We substitute this in Eq. (2) and collect the zeroth and
first order terms in Pef . At O(1), we get ψ(0) = 1/4π as 2. Numerical solution valid for arbitrary Pef . Now we
the isotropic microstructure. At O(Pef ), we have detail the numerical solution of Eq. (2), which uses the
n m
decomposition of ψ as ∑∞ n=0 ∑m=−n Cn,m (p)Yn (p). We
∇p ⋅{[ṗ(0) + De Pea ṗA /8π] ψ(0) }−τ Dr ∇2p ψ(1) +ψ(1) = 0.
(1) substitute this expansion in Eq. (2) and use the proper-
ties of spherical harmonics to obtain
(35)
Here we denote the equation of motion (Eq. 1) as having 1 ∞ n
(1) (1)
three parts: ṗ = ṗ(0) +DePef ṗP +DePea ṗA , where sub- − + ∑ ∑ Cn,m [1 + τ Dr n(n + 1)] Ynm +
4π n=0 m=−n
scripts represent passive and active components. Upon ∞ n
(1)
substituting this in Eq. (35), we find that the ṗP term Pef ∑ ∑ Cn,m H(Ynm ) = 0. (42)
2 n=0 m=−n
only contributes at O(Pef ), and is thus neglected. Upon
simplification, we find that the microstructure at O(Pef ) Here H(Ynm ) is a collection of expressions obtained by
is governed by simplifying ∇p ⋅ (ṗψ) . These are represented in terms of
3 angular momentum operators for computational conve-
−τ Dr ∇2p ψ(1) +ψ(1) = (Λ+α1 DePea /8π)E∞ ∶ pp. (36) nience [54] (detailed in Supplementary Note 3). We take
4π
 12

an inner product with Yij on both sides of Eq. (42) and The last term in the above expression is negligible and
use the orthogonality property to obtain is generated from the non-Newtonian component of ΣF p.
Thus, the major modifications due to fluid elasticity scale
1 j
− ∫ Y dp + Ci,j [1 + τ Dr i(i + 1)] + as Pe2a , as also depicted in Fig. 5d.
4π S i
∞ n
Pef ∑ ∑ Cn,m ∫ Yij H(Ynm )dp = 0. (43)
n=0 m=−n S
ACKNOWLEDGEMENTS
Since the n = 0 mode is already known from normaliza-
tion, the above equations can be recasted as the follow- Support from the Alexander von Humboldt Founda-
ing linear system of equations where Ci,j (for i ≥ 1) is tion is gratefully acknowledged. We also acknowledge
unknown: financial support from the Collaborative Research Cen-
Ci,j [1 + τ Dr i(i + 1)] + ter 910 funded by the Deutsche Forschungsgemeinschaft.
100 n
S.N acknowledges support from the Swedish Research
Pef ∑ ∑ Cn,m ∫ Yij H(Ynm )dp = Council (under grant no. 638-2013-9243) and Nordita,
n=1 m=−n S which is partially supported by NordForsk. A.C thanks
Jonas Einarsson for his advice on the effective equation
−Pef C0,0 ∫ Yij H(Y00 )dp., (44)
S of motion.
To find Ci,j coefficients, the above equations are solved
using the Clebsch-Gordon coefficient formulation, where
AUTHOR CONTRIBUTIONS
first 100 modes in n are used. We note that although the
numerical approach is valid for arbitrary Pef , there is
an indirect upper bound of weak viscoelasticity that we A.C. and H.S. planned the research. A.C performed
must adhere to, as we are using the SOF model. In non- the calculations. A.C., S.N., and H.S. analyzed the re-
dimensional terms this bound is: De Pef < 1. Thus, for sults and wrote the manuscript.
De = 0.1, we explore the results within an upper bound
of Pef = 10.
CONFLICTS OF INTEREST
Rheology. First we show the expanded version of the
flow-induced component of the particle stress
There are no conflicts to declare.
ΣF 3 ∇
p = −µf nl Aγ̇ a2 /2 (45)
where ai is a shorthand notation for the ensemble of
orientation moment of ith order: ⟨p⊗i ⟩. Expanding the
upper-convected derivative, we obtain
∇
a2 = (Λ − 1) (E∞ ⋅ a2 + a2 ⋅ E∞ ) − 2Λa4 ∶ E∞
+ De[Pef β1 (E∞ ⋅ a4 ∶ E∞ − 2E∞ ∶ a6 ∶ E∞ + E∞ ∶ a4 ⋅ E∞ )
+ Pef β2 (O∞ ⋅ a4 ∶ E∞ − E∞ ∶ a4 ⋅ O∞ )
+ Pea α1 (E∞ ⋅ a2 − 2a4 ∶ E∞ + a2 ⋅ E∞ )
+ Pea α2 (O∞ ⋅ a2 − a2 ⋅ O∞ ) ]. (46)

This ΣF p matches with the Newtonian bulk stress for

De = 0 [53]. For near-equilibrium results (Pef ≪ 1), the
above expression is substituted in Eq. (5) and orientation
distribution from Eq. (41) is used to perform the ensem-
ble integral. Since the passive viscoelastic contribution
is O(Pe2f ), it does not contribute at the O(Pef ) calcu-
lations. The viscosity relates to the deviatoric particle-
induced stress as ϕµp = Σp xy /γ̇, where ϕ = na3 is the
volume fraction. This viscometeric relation can be sim-
plified to obtain the near-equilibrium viscosity ratio
µp 4A(5 − 3Λ) 48Aτ Dr − Pea /π De Pea α1
= + (Λ + )
µf 15 5 (1 + 6τ Dr ) 8π
De Pea Aα1
− . (47)
10π
 1

Supplementary material for: “Orientational dynamics and rheology of active

suspensions in weakly viscoelastic flows”

I. COEFFICIENTS OF THE MULTIPOLE EXPANSION

The tensorial coefficients of the spheroidal multipoles are

pajklm = (pj pk − δjk /3)(pl pm − δlm /3), (S1a)

pbjklm = pj pl δkm + pj pm δkl + pk pl δjm + pk pm δjl − 4pj pk pl pm , (S1b)

pcjklm = −δjk δlm + δjl δkm + δjm δkl + pl pm δjk + pj pk δlm − pj pm δkl − pk pm δjl − pj pl δkm − pk pl δjm + pj pk pl pm . (S1c)

The other scalar coefficients are

√ √
R λ2 − 1 R λ2 − 1(λ2 + 1) R (λ2 − 1)3/2
A = , B = , C = , (S2a)
4(C − λ3 + λ) 4(C + λ3 − λ − 2Cλ2 ) 4(C + λ3 − λ − 2Cλ2 )

(λ2 − 1)3/2 (λ2 − 1)3/2 (Cλ + λ4 − 3λ2 + 2) (λ2 − 1)3/2

AS = , BS
= , C S
= .
4(C − 3λ3 + 3λ + 2Cλ2 ) 8(C + λ3 − λ − 2Cλ2 )(−3Cλ + λ4 + λ2 − 2) 2(3C + λ3 + 2λ5 − 7λ3 + 5λ)
(S2b)
√ √
Here C = λ2 − 1 coth−1 [λ/ λ2 − 1].

II. FURTHER RESULTS OF ORIENTATION DYNAMICS

A. Coefficients of equation of motion

Fig. S1 shows the coefficients of the analytical equation of motion. They vary weakly with aspect ratio λ and the
viscometric parameter δ, which typically lies between −0.7 and −0.5. Since the coefficients vary weakly with λ and
the integration time is very large for λ > 10, we consider aspect ratios between 5 and 10.

(a) (b)
1.5 𝛽"
𝛼!

1.0

0.5 𝛽!
𝛼"
0
5 6 7 8 9 10
𝜆 𝜆

Fig. S1. Coefficients of equation of motion (Eq. 1 in main text). αi represent active coefficients and βi are passive coefficients.

B. Alignment angle

Fig. S2 shows the comparison of angles of alignment obtained by integrating the analytical equation of motion
(Eq. 1 of main text) and those from the eigenvalue analysis. Note that the phase diagram has a weak dependency on
aspect ratio.
 2

3
4

𝜙!" 2

4 4

-4 -2 0 2
𝜎

Fig. S2. Comparison of alignment angles obtained by integrating the full equation of motion (dots) with those found using
analytical approximation (line). Blue: λ = 5, red: λ = 10. Vertical dashed lines separate the regions of alignment. Other
parameters: δ = −0.6, Wi = 0.1 .

C. Passive spheroid in weakly viscoelastic shear flow

The non-dimensional equation of motion for the orbits is

ṗ = (I − pp) ⋅ (E∞ ⋅ p) [Λ + Wi σα1 + Wi β1 E∞ ∶ pp] + Ω∞ × p [1 + Wi σα2 + Wi β2 E∞ ∶ pp] . (S3)
This can be written in the trigonometric form as
Λ
θ̇ = sin 2θ sin 2φ + Wi β1 sin3 θ cos θ sin2 φ cos2 φ (S4a)
4

1 Wi
φ̇ = − [λ2 sin2 φ + cos2 φ] + [β1 cos(2φ) − β2 ] sin2 θ sin 2φ. (S4b)
λ2 +1 4
To quantify the log-rolling and compare it with Brunn [37], we use the following transformation [59]:

⎛ Cλ ⎞ tan φ
θ = arctan √ , τ = arctan ( ). (S5)
⎝ λ2 cos2 φ + sin2 φ ⎠ λ

Here C represents the orbit constant (C = 0 and C = ∞ being the log-rolling and shear-plane rotation orbits, respec-
tively) and τ is essentially the time coordinate that is scaled with T /2π, where T is the Jeffery orbit. Using the above
relation, we obtain the evolution equation of the orbit constant as
[C 3 (β1 λ2 − β2 λ2 + β1 + β2 ) sin2 τ cos2 τ ]
Ċ(t) = Wi ,
2 (λ2 C 2 cos2 τ + C 2 sin2 τ + 1)
λ ⎡ sin τ cos τ (2β1 λ2 C 2 cos2 τ + β1 + β2 ) 1 ⎤
⎢ ⎥
τ̇ (t) = + Wi ⎢ − (β 1 + β 2 ) sin 2τ ⎥. (S6)
1 + λ2 ⎢ 2C 2 (λ2 cos2 τ + sin2 τ ) + 2 4 ⎥
⎣ ⎦
We now compare our result with Brunn [37] who, for a rigid tri-dumbbell, showed that the particle slowly drifts
towards the log-rolling orbit. We show the match between orbit drift from Eq. (S6) and Eq. (4.17) from Brunn in
Fig. S3. Note that over one period the particle’s orbit declines in a wiggly manner. Interestingly, these wiggles have
a physical significance and can be characterized as orbit reduction with two distinct plateaus. Fig. S3b shows that
the long plateau (around t/T = 1/4) occurs first and is the state where particle is in the vicinity of flow-vorticity plane
(where it slows down and spends most of its time). Once the particle exits the alignment state, it flips quickly to the
other side (−x axis). This flip generates the short plateau around t/T = 1/2.
.

III. DETAILS OF KINETIC MODEL

Here we provide full expressions and further details involved in the computational steps of kinetic model. The
spectral decomposition of ψ is carried out as
∞ n
ψ(θ, φ) = ∑ ∑ Cn,m (p)Ynm (p). (S7)
n=0 m=−n
 3

(a) 1.0 (b) 1.0

Our work
0.8
Brunn 0.9
C 0.6 C
C0 0.4 C0 0.8

0.2 0.7
0 2 4 6 8 10 0 0.2 0.4 0.6 0.8 1.0
t/T t/T

Fig. S3. Comparison with Brunn [37] for the variation of the orbit constant C/C0 , where C0 is the orbit constant for t = 0.
(a) Variation over 10 periods. (b) Variation over one period. Other parameters: λ = 5, T = 2π λ λ+1 , δ = −0.6, Wi = 0.1.
2

Substituting this in the Smoluchowski equation of main text and using the properties of spherical harmonics yields

n n
1 ∞ ∞
− + ∑ ∑ Cn,m [1 + τ Dr n(n + 1)] Ynm + Pef ∑ ∑ Cn,m H(Ynm ) = 0. (S8)
4π n=0 m=−n n=0 m=−n

Here H(Ynm ) is a term obtained by simplifying ∇p ⋅ (ṗψ) and is represented in terms of angular momentum operators
for computational convenience [54].

i Lz iΛ
H(Ynm ) = [− + (3iYnm sin2 (θ) sin(2φ) + Ly sin(2θ) sin(φ) + Lz cos(2θ) sin2 (φ) + Lz cos2 (φ))] +
2 2
i Pea
De{ [α1 (3iYnm sin2 (θ) sin(2φ) + Ly sin(2θ) sin(φ) + Lz cos(2θ) sin2 (φ) + Lz cos2 (φ)) − α2 Lz ] +
16π
i Pef
[β1 sin2 (θ)( − iYnm (3 + 5 cos(2θ)) − 10iYnm sin2 (θ) cos(4φ) + 2Ly sin(2θ) cos(φ) − 2Ly sin(2θ) cos(3φ)+
16
2Lz sin2 (θ) sin(4φ) + 2Lz cos(2θ) sin(2φ) + 2Lz sin(2φ)) + β2 sin2 (θ) (8iYnm cos(2φ) − 4Lz sin(2φ)) ]}. (S9)

Here, the angular momentum operators (L) are shorthand notation for the following gradients on harmonic Ynm :

∂Ynm ∂Y m ∂Ynm
Ly = Ly ∣Ynm ⟩ = −i cos φ + i cot θ sin φ n and Lz = Lz ∣Ynm ⟩ = −i . (S10)
∂θ ∂φ ∂φ

The trigonometric functions in the Eq. (S9) are converted into spherical harmonics. For example:

√
2π −1 1 √ √
sin θ cos θ sin φ = i (Y + Y21 ), cos2 θ = (2 4π/5Y20 + 4πY00 ) . (S11)
15 2 3
 4

Using such transformations, Eq. (S9) yields

√ √ √ √
√ 2π 2π 2π −2 2π 2
H(Ynm ) = [−i πLz Y00 + Λ(− Ly Y2 −
−1 1
Ly Y2 + i m
Y2 (Lz − 3Yn ) + i Y (Lz + 3Ynm )
15 15 15 15 2
√
1 √ 2 π
+ i πLz Y00 + i Lz Y20 )] +
3 3 5
α1Ly Y −1 α1Ly Y 1 iLz Y00 (α1 − 3α2) iα1Lz Y2−2 iα1Lz Y20 iα1Lz Y22
De{Pea [− √ 2 − √ 2 + √ + √ + √ + √ +
4 30π 4 30π 24 π 4 30π 12 5π 4 30π
√ √
1 3 m −2 1 3
− i α1Yn Y2 + i α1Ynm Y22 ] +
4 10π 4 10π
√ √ √ √ √
1 2π 1 2π 1 π 1 π 1 π
Pef [ i β1 Ly Y2 − i
−1
β1 Ly Y21 − i β1 Ly Y4−3 − i β1 Ly Y4−1 + i β1 Ly Y41 +
7 15 7 15 3 35 21 5 21 5
√ √ √ √
1 π 1 π 1 π 1 2π
i 3
β1 Ly Y4 − Lz Y2 (β1 − 7β2 ) +
−2 2
Lz Y2 (β1 − 7β2 ) − β1 Y4−4 (Lz − 5Ynm )+
3 35 7 30 7 30 3 35
√ √ √ √
1 2π m 1 2π 1 2π 2 π 2√
4
β1 Y4 (Lz + 5Yn ) − β1 Lz Y4 +
−2 2
β1 Lz Y4 + β1 Ynm Y20 − πβ1 Ynm Y40 +
3 35 21 5 21 5 7 5 21
√ √
2π m −2 2π
− β2 Yn Y2 − β2 Ynm Y22 ]} (S12)
15 15

The angular momentum operators are further simplified using the following relations [75]

1√ 1√
Ly = Ly ∣Ynm ⟩ = n(n + 1) − m(m + 1)Ynm+1 − n(n + 1) − m(m − 1)Ynm−1 and Lz = Lz ∣Ynm ⟩ = mYnm . (S13)
2 2
We substitute above relations in Eq. (S12) to simplify H(Ynm ) and use that in Eq. (S8). Next, we take an inner
product with Yij on both sides of Eq. (S8) and use the orthogonality property to obtain
n
1 j
∞
j m
− ∫ Yi dp + Ci,j [1 + τ Dr i(i + 1)] + Pef ∑ ∑ Cn,m ∫ Yi H(Yn )dp = 0. (S14)
4π S n=0 m=−n S

√ √
The first term is only non-zero for the zeroth mode (i = 0), which yields −1/ 4π. Since the C0,0 = 1/ 4π is already
known from normalization, the first term cancels the i = 0 mode of second term. Using these identities, Eq. (14) can
be recasted as the following linear system of equations in terms of Ci,j :
100 n
Ci,j [1 + τ Dr i(i + 1)] + Pef ∑ ∑ Cn,m ∫ Yij H(Ynm )dp = −Pef C0,0 ∫ Yij H(Y00 )dp, for i ≥ 1, (S15)
n=1 m=−n S S

where the right hand side is known. The two integral terms in the above set of equations contain the inner product of
three spherical harmonics. These are represented in terms of Clebsch-Gordon coefficients that exploit the orthogonal
relations between harmonics for computational ease [75]:
¿
π 2π Á (2a + 1)(2n + 1) a n N a n N
∫ ∫ YNM Yab Ynm sin θ dφdθ = Á
À [ ][ ]. (S16)
0 0 4π(2N + 1) 0 0 0 b m M

The bracket terms here are the Clebsch-Gordan coefficients inbuilt in Mathematica 13.
skip
Below we show the match between our results (obtained from the above kinetic model) in the limit De = 0 and
those obtained by Saintillan [24, Fig. 5a, τ̃ = 1] .

REFERENCES

[1] M. C. Marchetti, J. F. Joanny, S. Ramaswamy, T. B. Liverpool, J. Prost, M. Rao, and R. A. Simha, Rev. Mod. Phys. 85,
1143 (2013).
 5

0.3
0.2

μp
μf
0.1
0.0
100 101 102
Pef

Fig. S4. Comparing the shear viscosity as a function of Pef from our analysis in the Newtonian fluid limit with that of
Saintillan [24]. for ∣Pea∣
8πA
= 1, Λ = 1, τ = 1, Dr = 0.1. Lines (blue is puller, red is pusher) indicate our viscosity ratio and dots
correspond to Saintillan [24], whose viscosity is defined here with respect to the particle volume fraction ϕ = na3 . Saintillan
[24] used ϕ = nl3 A.

[2] A. Zöttl and H. Stark, Journal of Physics: Condensed Matter 28, 253001 (2016).
[3] E. Lauga, The fluid dynamics of cell motility, Vol. 62 (Cambridge University Press, 2020).
[4] H. C. Berg and D. A. Brown, nature 239, 500 (1972).
[5] M. Demir and H. Salman, Biophysical journal 103, 1683 (2012).
[6] H. Stark, Accounts of Chemical Research 51, 2681 (2018).
[7] A. J. Mathijssen, N. Figueroa-Morales, G. Junot, É. Clément, A. Lindner, and A. Zöttl, Nat. Comm. 10, 1 (2019).
[8] G. Jing, A. Zöttl, É. Clément, and A. Lindner, Science advances 6, eabb2012 (2020).
[9] V. S. Doan, P. Saingam, T. Yan, and S. Shin, ACS nano 14, 14219 (2020).
[10] A. Ramamonjy, J. Dervaux, and P. Brunet, Physical Review Letters 128, 258101 (2022).
[11] S. S. Suarez and A. Pacey, Hum. Reprod. Update 12, 23 (2006).
[12] A. Zöttl and H. Stark, Phys. Rev. Lett. 108, 218104 (2012).
[13] J. Sznitman and P. E. Arratia, Complex Fluids in Biological Systems (Springer, 2015) pp. 245–281.
[14] R. Shogren, T. A. Gerken, and N. Jentoft, Biochemistry 28, 5525 (1989).
[15] S. K. Lai, Y.-Y. Wang, D. Wirtz, and J. Hanes, Advanced drug delivery reviews 61, 86 (2009).
[16] A. J. Mathijssen, T. N. Shendruk, J. M. Yeomans, and A. Doostmohammadi, Phys. Rev. Lett. 116, 028104 (2016).
[17] A. Choudhary and H. Stark, Soft Matter 18, 48 (2022).
[18] A. Sokolov and I. S. Aranson, Physical review letters 103, 148101 (2009).
[19] J. Gachelin, G. Mino, H. Berthet, A. Lindner, A. Rousselet, and É. Clément, Physical review letters 110, 268103 (2013).
[20] A. G. McDonnell, T. C. Gopesh, J. Lo, M. O’Bryan, L. Y. Yeo, J. R. Friend, and R. Prabhakar, Soft Matter 11, 4658
(2015).
[21] H. M. López, J. Gachelin, C. Douarche, H. Auradou, and E. Clément, Phys. Rev. Lett. 115, 028301 (2015).
[22] Y. Hatwalne, S. Ramaswamy, M. Rao, and R. A. Simha, Phys. Rev. Lett. 92, 118101 (2004).
[23] B. M. Haines, A. Sokolov, I. S. Aranson, L. Berlyand, and D. A. Karpeev, Physical Review E 80, 041922 (2009).
[24] D. Saintillan, Exp. Mech. 50, 1275 (2010).
[25] G. B. Jeffery, Proceedings of the Royal Society of London. 102, 161 (1922).
[26] S. Rafaı̈, L. Jibuti, and P. Peyla, Phys. Rev. Lett. 104, 098102 (2010).
[27] G. Li and A. M. Ardekani, Phys. Rev. Lett. 117, 118001 (2016).
[28] G. Li, E. Lauga, and A. M. Ardekani, Journal of Non-Newtonian Fluid Mechanics 297, 104655 (2021).
[29] S. E. Spagnolie and P. T. Underhill, arXiv preprint arXiv:2208.03537 (2022).
[30] A. Patteson, A. Gopinath, M. Goulian, and P. Arratia, Sci. rep. 5, 15761 (2015).
[31] A. Zöttl and J. M. Yeomans, Nature Physics 15, 554 (2019).
[32] S. Kamdar, S. Shin, P. Leishangthem, L. F. Francis, X. Xu, and X. Cheng, Nature 603, 819 (2022).
[33] S. Liu, S. Shankar, M. C. Marchetti, and Y. Wu, Nature 590, 80 (2021).
[34] R. B. Bird, R. C. Armstrong, and O. Hassager, Dynamics of polymeric liquids. Vol. 1: Fluid mechanics (1987).
[35] D. F. James, Annual Review of Fluid Mechanics 41, 129 (2009).
[36] L. Leal, J. Fluid. Mech. 69, 305 (1975).
[37] P. Brunn, J. Fluid Mech. 82, 529 (1977).
[38] E. Hinch and L. Leal, J. Fluid. Mech. 71, 481 (1975).
[39] A. T. Chwang and T. Y.-T. Wu, Journal of Fluid mechanics 67, 787 (1975).
[40] S. E. Spagnolie and E. Lauga, J. Fluid Mech. 700, 105–147 (2012).
[41] A. P. Berke, L. Turner, H. C. Berg, and E. Lauga, Phys. Rev. Lett. 101, 038102 (2008).
[42] K. Drescher, R. E. Goldstein, N. Michel, M. Polin, and I. Tuval, Phys. Rev. Lett. 105, 168101 (2010).
[43] J. Einarsson, F. Candelier, F. Lundell, J. Angilella, and B. Mehlig, Physics of Fluids 27, 063301 (2015).
[44] F. Gauthier, H. Goldsmith, and S. Mason, Rheo.l Acta 10, 344 (1971).
[45] G. D’Avino and P. L. Maffettone, Journal of Non-Newtonian Fluid Mechanics 215, 80 (2015).
[46] H. Berg, Random Walks in Biology. Princeton Univ. Press, Princeton, NJ (1983).
[47] E. Lauga, Ann. Rev. Fluid Mech. 48, 105 (2016).
 6

[48] T. C. Adhyapak and H. Stark, Soft Matter 12, 5621 (2016).

[49] S. Nambiar, P. Nott, and G. Subramanian, Journal of Fluid Mechanics 812, 41 (2017).
[50] M. Doi, S. F. Edwards, and S. F. Edwards, The theory of polymer dynamics, Vol. 73 (Oxford University Press, 1988).
[51] C. Cohen, B. Chung, and W. Stasiak, Rheo. Acta 26, 217 (1987).
[52] S. Chattopadhyay, R. Moldovan, C. Yeung, and X. Wu, Proc. Natl. Acad. Sci. 103, 13712 (2006).
[53] E. Hinch and L. Leal, J. Fluid. Mech. 76, 187 (1976).
[54] S. B. Chen and D. L. Koch, Physics of Fluids 8, 2792 (1996).
[55] E. Hinch and L. Leal, Journal of Fluid Mechanics 52, 683 (1972).
[56] A. J. Wolfe, M. P. Conley, and H. C. Berg, Proceedings of the National Academy of Sciences 85, 6711 (1988).
[57] B. Chung and C. Cohen, J. Nonnewton. Fluid Mech. 25, 289 (1987).
[58] Y. Iso, D. L. Koch, and C. Cohen, J. Nonnewton. Fluid Mech. 62, 115 (1996).
[59] L. Leal and E. Hinch, Journal of Fluid Mechanics 46, 685 (1971).
[60] J. Elgeti, R. G. Winkler, and G. Gompper, Reports on progress in physics 78, 056601 (2015).
[61] G. Junot, N. Figueroa-Morales, T. Darnige, A. Lindner, R. Soto, H. Auradou, and E. Clément, EPL (Europhysics Letters)
126, 44003 (2019).
[62] A. Brown, S. Clarke, P. Convert, and A. Rennie, Journal of rheology 44, 221 (2000).
[63] A. Einstein, Ann. Phys. 19, 289 (1906).
[64] J. Férec, E. Bertevas, B. Khoo, G. Ausias, and N. Phan-Thien, Journal of Non-Newtonian Fluid Mechanics 239, 62
(2017).
[65] R. B. Bird, C. F. Curtiss, R. C. Armstrong, and O. Hassager, Dynamics of polymeric liquids, volume 2: Kinetic theory
(Wiley, 1987).
[66] L. Ren, N. Nama, J. M. McNeill, F. Soto, Z. Yan, W. Liu, W. Wang, J. Wang, and T. E. Mallouk, Sci. Adv. 5, 3084
(2019).
[67] A. Aghakhani, O. Yasa, P. Wrede, and M. Sitti, Proc. Natl. Acad. Sci. 117, 3469 (2020).
[68] Y. Bozorgi and P. T. Underhill, Physical Review E 84, 061901 (2011).
[69] A. M. Ardekani and E. Gore, Phys. Rev. E 85, 056309 (2012).
[70] N. Narinder, C. Bechinger, and J. R. Gomez-Solano, Phys. Rev. Lett. 121, 078003 (2018).
[71] S. A. Abtahi and G. J. Elfring, Physics of Fluids 31, 103106 (2019).
[72] G. J. Elfring and E. Lauga, in Complex fluids in biological systems (Springer, 2015) pp. 283–317.
[73] F. P. Bretherton, J. Fluid Mech. 14, 284 (1962).
[74] S. H. Strogatz, Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering (CRC
Press, 2018).
[75] G. B. Arfken and H. J. Weber, Mathematical methods for physicists (American Association of Physics Teachers, 1999).