Turbulence in a network of rigid fibers

Stefano Olivieri1,2,3 , Assad Akoush4 , Luca Brandt5 , Marco E. Rosti1 and Andrea Mazzino2,3
1
Complex Fluids and Flows Unit, Okinawa Institute of Science and Technology Graduate University,
1919-1 Tancha, Onna-son, Okinawa 904-0495, Japan
2
Department of Civil, Chemical and Environmental Engineering (DICCA),
University of Genova, Via Montallegro 1, 16145, Genova (Italy)
3
INFN, Genova Section, Via Montallegro 1, 16145, Genova (Italy)
4
Department of Mechanical Engineering, Faculty of Engineering and Architecture,
American University of Beirut, P.O. Box 11-0236, Riad El Solh, Beirut 1107 2020 (Lebanon)
arXiv:2005.00442v1 [physics.flu-dyn] 1 May 2020

5
Linné FLOW Centre and SeRC, Department of Engineering Mechanics,
KTH Royal Institute of Technology, Stockholm, Sweden
(Dated: May 4, 2020)
The effect of a network of fixed rigid fibers on fluid flow is investigated by means of three-
dimensional direct numerical simulations using an immersed boundary method for the fluid-structure
coupling. Different flows are considered (i.e., cellular, parallel and homogeneous isotropic turbulent
flow) in order to identify the modification of the classic energy budget occurring within canopies
or fibrous media, as well as particle-laden flows. First, we investigate the stabilizing effect of the
network on the Arnold-Beltrami-Childress (ABC) cellular flow, showing that, the steady configura-
tion obtained for a sufficiently large fiber concentration mimics the single-phase stable solution at
a lower Reynolds number. Focusing on the large-scale dynamics, the effect of the drag exerted by
the network on the flow can be effectively modelled by means of a Darcy’s friction term. For the
latter, we propose a phenomenological expression that is corroborated when extending our analysis
to the Kolmogorov parallel flow and homogeneous isotropic turbulence. Furthermore, we exam-
ine the overall energy distribution across the various scales of motion, highlighting the presence of
small-scale activity with a peak in the energy spectra occurring at the wavenumber corresponding
to the network spacing.

I. INTRODUCTION

Scale separation is one of the most successful concepts exploited in a wide range of physical domains to derive
large-scale dynamical descriptions where the small-scale dynamics only appear via effective parameters. The best
known examples come from the solid state physics and kinetic theory. In the first field, the scale separation between
the time scale associated to the motion of atomic nuclei and that of the electrons in a molecule allows one to treat
them separately. As a result, the ion-ion interaction can be described by effective potentials on account of the motion
of electrons. This is the essence of the well-known Born-Oppenheimer approximation.
In the gas kinetic theory, when the scales of interest are much larger than the mean free path (i.e., when the so-called
Knudsen number is small), hydrodynamic equations can be derived where only macroscopic scales are involved, with
the smallest involved only via effective parameters.
Scale separation also constitutes the foundation for the emergence of effective parameters in the infra-red limit of
fluid turbulence and turbulent transport: eddy-viscosity [1–3] and eddy-diffusivity [4–9] are well-known examples of
effective parameters accounting for the non resolved small-scale dynamics.
Our aim here is to investigate how turbulence is modified when interacting with a network of slender rigid fibers
of length within the turbulence inertial range. Such a system can be seen as a rough model for quantitatively
assessing how turbulence is modified by interacting with a canopy. This multiscale problem will be tackled by
high-resolution DNS complemented by a state-of-the-art immersed boundary method (IBM) to fully resolve the
fluid-structure interaction problem. Thanks to the full resolution of the whole coupled system, both the small-scale
dynamics and the effective large-scale dynamics will be investigated. The latter will clearly emerge in the form of an
effective Darcy equation when the flow field is observed on scales much larger than the typical correlation length of
the solid network.
One of the motivations of our study comes from boundary-layer meteorology where the interaction of wind with
plant or urban canopies is known to cause modifications in the momentum and heat fluxes and velocity profiles [10].
Such features can influence the transport and mixing properties within the canopy, consequently altering ecological
mechanisms such as carbon dioxide exchange [11]. Indeed, a relevant number of studies have been devoted to un-
derstand the underlying mechanisms by means of laboratory or in-situ measurements [12–15], as well as numerical
simulations [16–18]. Detailed reviews on how turbulence is modified in plant canopies can be found in Refs. [11, 19, 20].
Finnigan [19] highlighted, in particular, how the classical turbulent scenario, well explained by Kolmogorov theory [5],
can be substantially modified in the case of canopy flow. This is caused by the presence of canopy elements (e.g. twigs
 2

FIG. 1. Left: homogeneous isotropic turbulent flow (the colormap showing one velocity component). Right: network of
N = 63 fibers of length c/L = (2π)−1 placed in the same triperiodic fluid domain.

and leaves) exerting both viscous and pressure drag. According to Finnigan [19], the former has an overall dissipative
effect, while the latter is responsible for generating wakes behind canopy elements, so that part of the large-scale
kinetic energy is converted into smaller-scale kinetic energy (i.e., at higher wavenumbers). Such energy transfer was
thus labeled as the ‘spectral shortcut’ mechanism, which is believed to explain why the classical inertial subrange
scaling E(k) ∼ k −5/3 does not hold within canopies [10].
Such explanation, however, has been so far rather empirical and qualitative, lacking of a thorough analysis in a
more quantitative framework. To gain a deeper comprehension, in the present work, the problem will be tackled
in a more fundamental way, focusing on a relatively simple yet representative model: the fluid-solid interaction
between a network of rigid fibers and several three-dimensional flows, i.e. (i) ABC cellular flow, (ii) Kolmogorov
parallel flow, (iii) homogeneous isotropic turbulence. Our three models aim to be an idealized but sufficiently general
representation of both laminar flows with open and closed streamlines and a fully developed turbulent flow. In
particular, the choice of the cellular and parallel flows will enable us to explore how the presence of fibers alters
the stability properties compared to the single-phase case (i.e., without fibers), which has been already extensively
investigated [21, 22]. The insights obtained in these two configurations will then be confirmed in the framework
of homogeneous isotropic turbulence. Additionally, we highlight that this kind of study has relevance also for the
understanding of the interaction between fluid flows and dispersed fiber-like objects with large inertia, a topic of
interest for many environmental/biological and industrial problems, such as aerosol deposition in the production of
composite materials [23].
The rest of the paper is structured as follows: Sec. II introduces the physical model and governing equations, along
with describing how these are solved numerically; Sec. III presents the results of our analysis and the main conclusions
are drawn in Sec. IV.

II. PHYSICAL MODEL AND GOVERNING EQUATIONS

We consider a cubic domain with side L and periodic boundary conditions, with an ensemble of approximately
one-dimensional fibers immersed within, as shown in Fig. 1. Each fiber composing the network has a length c and
diameter d such that the aspect ratio c/d ≫ 1. An integer number of fibers is placed along each direction with their
centroids evenly spaced (one case of randomly spaced fibers will also be considered), while the orientation is always
uniformly random distributed. The fiber concentration can be characterized by the number density n = N/V , where
N is the number of fibers and V = L3 is the total volume of the fluid domain.
The fluid motion is governed by the incompressible Navier-Stokes equations
∂t u + u · ∂u = −∂p/ρ0 + ν∂ 2 u + f , (1)

∂ · u = 0, (2)
where u = (u, v, w) is the fluid velocity, p the pressure, ρ0 and ν the density and kinematic viscosity of the fluid and f
is the external volume forcing. The latter consists of two contributions, i.e. f = fFOR + fFIB : (i) fFOR is a body force
 3

TABLE I. List of the cases for ABC cellular flows with a network of evenly spaced N fibers of length c. D here is the value
measured from the decay of the energy spectrum.
N c/L n D (measured) Re eff State
0 – 0 – 130 chaotic
33 (2π)−1 0.11 – 53.9 chaotic
43 (2π)−1 0.26 – 29.9 chaotic
53 (2π)−1 0.50 – 17.2 pseudoperiodic
63 (2π)−1 0.87 0.11 10.5 stable
73 (2π)−1 1.38 0.16 6.8 stable
83 (2π)−1 2.06 0.19 4.6 stable
63 (4π)−1 0.87 – 16.0 pseudoperiodic
73 (4π)−1 1.38 0.10 10.6 stable
83 (4π)−1 2.06 0.14 7.3 stable
103 (4π)−1 4.03 0.23 3.8 stable

used to generate and sustain the desired flow, while (ii) fFIB is the fluid-structure coupling term used to account for
the presence of the fibers by an immersed boundary method (IBM).
Specifically, we employ the same numerical procedure already used for moving and deforming filaments in laminar
or turbulent flows in Refs. [24–27], to which the reader is referred to for further information. The implementation relies
on a finite difference, fractional step method on a staggered grid with fully explicit second-order central-differencing
scheme in space and third-order Runge-Kutta scheme in time. Additionally, the Poisson equation enforcing the
incompressibility constraint is solved using the Fast Fourier Transform.
The fluid-structure coupling force fFIB is computed following the method by Huang et al. [28] and later modified
by Banaei et al. [26]. In particular, the Lagrangian force is first evaluated at each material point belonging to the
fibers to enforce the no-slip condition U(X(s, t), t) = Ẋ = 0 as

F(s, t) = κ (Ẋ − U), (3)

where κ is a large negative constant [28] and

Z
U(X(s, t), t) = u(x, t)δ(x − X(s, t)) dx (4)

is the interpolated fluid velocity at the position X = X(s, t) of the material point belonging to the fiber, as a function
of the curvilinear coordinate s and time t. A spreading is thus performed over the surrounding Eulerian points,
yielding the volumetric forcing acting on the flow

Z
fFIB (x, t) = F(s, t)δ(x − X(s, t)) ds. (5)

Both the interpolation and spreading feature the Dirac operator, which is discretised by a regularized δ; here, we
employ the function proposed by Roma et al. [29].
In our simulations, we choose a domain size L = 2π and discretize into a Cartesian grid using 64 points per side,
with periodic boundary conditions applied in all directions. The fiber elements are discretized into (NL − 1) segments
with spatial resolution ∆s = c/(NL − 1), NL being the number of Lagrangian points. Here we use NL = 11 points, so
that the Lagrangian spacing ∆s is approximately equal to the Eulerian grid size ∆x. We verified that the variation
of the results is negligible when doubling both the Eulerian and Lagrangian resolutions. As for the timestep, we use
∆t = 5 × 10−5 , after assessing convergence.
The described procedure has been implemented and extensively validated in both laminar and turbulent flow
conditions: for more information the reader is referred to Rosti et al. [24, 25], Banaei et al. [26], Rosti and Brandt
[30].
 4

0.2
N=0 N = 33 N = 43 N = 53
0.15

2 hu i
1 2
0.1

0.05

0
400 600 800 1000 1200 1400 1600

FIG. 2. Time history of the mean fluid kinetic energy for different fiber concentrations in cellular ABC flow at Re = 130.

III. RESULTS

A. ABC cellular flow

To start, we consider the so-called Arnold-Beltrami-Childress (ABC) flow [31]:

u = A sin z + C cos y
v = B sin x + A cos z (6)
w = C sin y + B cos x

which is known to be a time-independent three-dimensional solution of both the Euler and Navier-Stokes equations,
under the forcing fFOR = ν(A sin z + C cos y, B sin z + A cos z, C sin y + B cos z), provided that the Reynolds number
Re ≡ ν −1 is sufficiently small [21, 31, 32]. The ABC flow is a special case of the Beltrami flows where the parameters
A, B, and C are real and the flow is periodic with respect to the Cartesian coordinates [33]. Despite its simple
analytical expression, the flow provides an example of Lagrangian chaos [31, 34]. Stability analyses on this system
have been carried out by several authors, see e.g. Refs. [21, 22], which can therefore be used as a reference to study
the effect of a network of fibers. In particular, choosing A = B = C = 1, the flow becomes unsteady for Re ≥ 13,
showing an oscillatory behavior for 13 < Re ≤ 20 and becoming chaotic for Re > 20 [22].
Here, we therefore fix Re = 130 (for which in the absence of the dispersed phase the system is known to be unstable)
and give to the initial condition a perturbation of finite amplitude at the same wavenumber of the forcing k = 1. As
a result, the solution given by Eq. (6) is lost in favour of an unsteady and chaotic regime [22]. Using this setting, we
conducted a parametric study by varying the number of fibers N and the fiber length c, as presented in Table I.
Fig. 2 reports the time history of the mean fluid kinetic energy hu2 i/2 (where h·i denotes spatial average over the
fluid domain) for cases with different concentration. Starting from the unstable flow without fibers (i.e. N = 0),
both the average value and the oscillation amplitude decrease for increasing N . Furthermore for sufficiently large
concentrations (N ≥ 53 ), we eventually reach a completely steady behavior. Thus, as expected, the presence of the
fiber network stabilizes the flow provided that the network is sufficiently dense of fibers.
This result is qualitatively confirmed by the three-dimensional visualizations of the velocity field reported in Fig. 3.
In the absence of the fiber network, the stable solution (left panel) given by Eq. (6) is obtained only if Re < Re cr ,
while we obtain the unsteady and chaotic solution for Re > Re cr (center panel). Let us now focus on the case with
fibers (right panel); as already pointed out, for a sufficiently dense network the resulting configuration is stable and
steady. Looking at the velocity field, two main features can be observed: (i) the presence of wakes around the network
elements, which can be associated to small-scale activity; (ii) the resemblance of the large-scale flow structure with
that of the (stable) ABC flow.
Such evidence suggests to separate these two aspects by means of a scale-by-scale analysis. We have therefore
computed the corresponding energy spectra, reported in Fig. 4a, from which it can be observed how the energy
distribution across the scales of motion is modified by the presence of the network. We observe that, the energy
associated to the large-scale/low-wavenumber components appears to decrease, while the small-scale/high-wavenumber
activity becomes more relevant, consistently with what previously noted from the field visualization. One can notice
that k = 1, the scale where the energy is introduced, remains always the dominant mode. For sufficiently high
concentrations however, a secondary peak is seen to emerge. The wavenumber associated to this local maximum can
 5

FIG. 3. Visualizations of one component of the velocity field for three-dimensional cellular ABC flows. Left: stable solution
given by Eq. (6); center: unstable flow at Re = 130 (at a given time instant); right: steady flow obtained in the presence of a
network of N = 63 evenly spaced fibers with c/L = (2π)−1 .

√
3
be identified as kc = 2π/ℓ = √ N , ℓ being the characteristic lengthscale associated with the spacing between the
network elements, i.e. ℓ = 2π/ 3 N . This scale is activated by virtue of the no-slip boundary condition imposed on
each fiber. It can be noted that the wavenumber associated with the fiber length, i.e. 2π/c, does not appear directly
involved, differently from what is typically claimed for turbulent flows within canopies [10, 19].
The structure of the energy spectra discussed above, where a certain scale separation occurs, suggests the possibility
of an effective description for the large-scale dynamics. To investigate this aspect, we look at how the low-wavenumber
components of the energy spectrum decay in time. Here we focus on the case N = 83 fibers, but similar findings
are obtained for all the cases with sufficiently dense networks. The initial condition is a given frame from the fully-
developed unstable configuration without fibers previously described. The time history of the first two components of
E(k = 1, 2) is shown in Fig. 5a. Except for the initial stages, the decay is substantially exponential for both modes, so
that it can be expressed as E(k, t) = E0 (k) exp(−βt) + E∞ (k), where E0 and E∞ are the initial and asymptotic value
(note that in the plot the spectrum is subtracted by the latter in order to highlight the exponential behavior), while
β is the characteristic time decay rate. From the energy balance, the governing equation for the energy spectrum can
be written as follows:
∂t E(k, t) = T (k, t) + V (k, t) + FFOR (k, t) + FFIB (k, t), (7)
with the various terms of the right-hand-side corresponding to the nonlinear energy transfer, the viscous dissipation,
the external flow forcing and the fluid-structure coupling, respectively; for the definition of these quantities, see
Appendix A. For the moment, let us neglect the nonlinear term (this assumption will be later justified).
From Fig. 5a, we can observe that the first and second spectral mode decay essentially at the same rate (in this
case β ≈ 0.4). This indicates that, when focusing on the large-scale dynamics, the effect of the network of fibers can
be modelled by means of a Darcy-like friction term FFIB (k, t) = −D E(k, t), where D represents the friction factor.
Indeed, the decay associated with such friction term turns out to be independent from the wavenumber k. Conversely,
for the viscous dissipation V (k, t) = −2νk 2 E(k, t) it scales as k 2 . Considering both contributions, the overall decay
rate is thus β = 2(D + νk 2 ). Testing this assumption against our DNS data, the effective Darcy friction is found to
be dominant compared with the viscous dissipation, resulting in the observed independence of β from the spectral
mode k.
We thus focus only on the largest scale, i.e. on the first wavenumber k = 1, and shift our attention back to physical
space, where the fluid-structure forcing can be modelled now as fFIB = −Du. Consequently, we can write a balance
between the resulting large-scale velocity field ueff (governed by the external forcing, the viscous and the Darcy terms)
and the single-phase solution of the ABC flow uABC (Eq. (6)) (governed only by the forcing and the viscous terms) as
Dueff − ν∂ 2 ueff + fFOR = −ν∂ 2 uABC + fFOR . (8)
Combining Eqs. (8) and (6), the large-scale velocity field ueff is thus simply obtained as
ν
ueff = uABC , (9)
ν +D
i.e., the large-scale flow is the same as the ABC solution, although with a reduced amplitude compared to the one
obtained without fibers. Note that, this is formally valid only when the nonlinear terms are small so that they can
be neglected in the balance.
 6

(a)

10−1

10−2

10−3

E(k)
10−4

10−5 N
10−6

10−7
1 10

(b)

101

100
2 2
i
E(k) α nL(c/L) 3

10−1

10−2
h

10−3

10−4
2
3

10−5
1 10

(c)

101

100
2 2
i
E(k) α nL(c/L) 3

10−1
k−3
10−2
h

10−3

10−4
2
3

10−5
0.1 1

k/kc

FIG. 4. (a) Energy spectra in ABC cellular flow at Re = 130 for different fiber concentrations (black: N = 0, yellow: N = 33 ,
blue: N = 43 , violet: N = 53 , green: N = 63 , light blue: N = 73 , orange: N = 83 , red: N = 103 ) and lengths (solid:
c/L = (2π)−1 , dashed: c/L = (4π)−1 , dotted: c/L = (2π)−1 with random positions). (b) Spectra of stabilized cases normalized
using our argument based on Eqs. (9) and (10). (c) Same but normalizing also the independent variable using the network
wavenumber.
 7

(a)
10−1
0.3
10−2
k=1
10−3

D
E(k,t) − E∞ (k)
exp(−
10−4 βt ) 0
0.005 0.04
10−5 k=2 ν nL(c/L)2/3

10−6
10−7
10−8
10−9
0 5 10 15 20 25 30
t

(b)
0.1

0.08

0.06

0.04
u

0.02

−0.02
0 1 2 3 4 5 6
z

FIG. 5. (a) Time history of the first and second mode of the energy spectrum for ABC flow at Re = 130 and network of
N = 83 fibers with c/L = (2π)−1 (solid blue line: k = 1, dashed blue line: k = 2, black dotted line: ∼ exp(−βt)); the inset
reports the Darcy friction coefficient measured from our DNS (filled squares: c/L = (2π)−1 , empty circles: c/L = (4π)−1 ) along
with its expression proposed in Eq. (10) (dashed line). (b) Example of velocity profiles (u-component along z-direction) from
the fully-resolved simulation (black solid line), the same but applying a large-scale filter (blue solid line) and that obtained
using Eq. (9).

To validate our argument, Fig. 5b reports a sample of the resulting velocity profile, which can be filtered to remove
the small-scale components retaining only the first modes. Comparing the latter with the profile given by Eq. (9),
where we highlight that D is measured from the decay of the energy spectrum, we find that the agreement is very
good, confirming the validity of our approach. Such procedure has been applied to the other five cases reported in
Table I for which a stable state is reported, with the aim of deducing an expression of D as a function of the main
parameters involved in the problem. From the numerical evidence, we find that the Darcy’s coefficient can be written
as
 c  23
D = α ν nL , (10)
L
where α ≈ 7 is a dimensionless factor that we have found empirically by fitting the expression to our data. This
is graphically shown in the inset of Fig. 5a, where the results obtained for networks with different N and/or c/L
collapse reasonably well. The form of Eq. (10) deserves some comments: i) both the fluid viscosity and the network
concentration enter linearly into the expression; ii) the box size L is used as the governing characteristic length instead
of c as representative of the elementary cell volume; iii) the dependence of D on the fiber length is found to be weaker
than linear. Note that, overall this is similar to the typical structure of the Darcy’s term used for porous media
models [35].
 8

(a) (b)

0.004 0.001
0.003
0.002 0.0005
0.001
k F(k)

k F(k)
0 0
−0.001
−0.002 −0.0005
−0.003
−0.004 −0.001
1 10 1 10

k k

FIG. 6. Spectral power balance according to Eq. (7), multiplied by k to improve the plot readibility, for cellular ABC flow at
Re = 130: (a) unstable case without fibers (N = 0), and (b) stabilized flow for a network of N = 103 fibers with c/L = (4π)−1 .
Black solid line: external forcing; red dashed: fluid-structure coupling; blue dotted: nonlinear term; violet dot-dashed: viscous
dissipation.

Using again the similarity with the stable ABC solution (Eq. (9)), we can evaluate an effective Reynolds number,
Re eff = (ν + D)−1 , on the basis of which the fluid flow can be characterized. To be consistent with the results of the
stability analysis for the classical ABC flow, Re eff . Re cr , where Re cr ≈ 13 is the aforementioned critical value for
the first instability. In Table I, we report the value of Re eff along with the observed state (e.g., stable or chaotic) for
each case: the correlation between the two is as expected, further confirming that the approach here proposed can
effectively model the presence of the network.
Moreover, the magnitude of the first mode of the energy spectrum can be derived from Eq. (9). Indeed, in the
classical ABC flow the latter is equal to 3/2A2 , while in the presence of fibers we have 3/2A′2 , with A′ = ν/(ν + D) ≈
[αnL(c/L)2/3 ]−1 (neglecting the viscous term contribution). Using this quantity, we normalize the energy spectra
(of those cases where Re eff < Re cr ) as shown in Fig. 4b, and show that the large-scale/low-wavenumber components
substantially overlap. Moreover, the energy spectra can be normalized also in the independent variable using the
network wavenumber kc discussed previously, as depicted in Fig. 4c where one can notice that all curves are collapsing
for k/kc ≥ 1, i.e. in the small-scale range. In this range the scaling resembles a power law ∼ k −3 , the fingerprint of a
regime having smooth fluctuations in space.
One important aspect has to be underlined regarding the limit of validity of Eq. (9): as we already stated, the
balance that we considered, between the effective friction and the external forcing, relies on the fact that the fiber
network is sufficiently dense so that the friction is large enough and the nonlinear terms in the Navier-Stokes equations
can be neglected (the latter will be discussed later together with Fig. 6b).
To complete the analysis, we consider again the spectral energy budget in Eq. (7) and compute from our numerical
data each term appearing on the right-hand-side, reported in Fig. 6 for the unstable flow and one case stabilized by
the fiber network. We first consider the single-phase case (Fig. 6a). As prescribed, the external forcing acts only on
the first mode. The energy input is balanced only partially by dissipation, while the remaining is transferred by the
nonlinear term to higher wavenumbers. Note that here the flow is chaotic but not properly turbulent, due to the
limited Re, so that no characteristic energy cascade can be observed. Nevertheless, although limited to few modes, a
certain proliferation of active scales of motion occurs (see again the black curve in Fig. 4a), up to the condition where
viscous dissipation becomes dominant.
Let us now move to the case with a network of N = 103 fibers with c/L = (4π)−1 (Fig. 6b). While the external forcing
has obviously the same structure as before, the energy input from the external forcing is reducing (compared to the
case without fibers) because of the reduced mean kinetic energy (as it was shown in Fig. 2). Remarkably, the nonlinear
term is found to be negligible for every k if compared to the other terms, thus justifying the assumption previously
made. Focusing on k = 1, it is evident how the large-scale dynamics is given by a balance between the forcing and
the fluid-structure coupling: the former injects energy in the system, while the latter subtracts it. Furthermore,
Fig. 6b also shows a positive peak for the fluid-structure coupling at k = kc that is substantially balanced by viscous
dissipation. The same behavior occurs at higher wavenumbers k > kc , although with the tendency for both terms
 9

to decrease with k. We also point out that all terms are vanishing over an intermediate range of wavenumbers,
approximately 2 ≤ k ≤ 6, indicating once again the nonlocal energy transfer operated by the fiber network.
To summarize, the analysis carried out on the cellular flow reveals the stabilizing effect of a sufficiently dense
network of fibers, which can be effectively described using a Darcy-like friction term when focusing on the larger scales
of motion. At smaller scales, a nonlocal energy transfer mechanism is responsible of a secondary peak in the energy
spectrum emerging at a lengthscale associated with the spacing between the network elements. Before concluding
this section, it is worth noticing that the whole analysis holds also when the fibers are arranged completely random.
In this case, as shown in Fig. 4 with dotted lines, the spectra still present the same features previously described
and the only difference is a broader secondary peak than in the ordered case, being the lengthscale associated with
the spacing between the network elements not uniquely defined anymore. In the following sections, we aim to assess
whether the proposed phenomenological model still works in different flows.

B. Kolmogorov flow

As a complementary case to the cellular flow previously examined, we now consider a parallel flow configuration.
In particular, we choose the so-called Kolmogorov flow, defined as [36]

u = cos y
v=0 (11)
w = 0,

where the streamlines are open and the only nonzero velocity component varies sinusoidally along the transverse
direction. To obtain the solution above (provided that Re < Re cr ), the external forcing in Eq. (1) is now expressed
as fFOR = −ν cos y ex . Similarly to the ABC flow, this configuration represents a prototype for the investigation of
hydrodynamic stability, both for Newtonian and non-Newtonian fluids √ [37–41]. In the single-phase case, the critical
value of the Reynolds number is found theoretically to be Re cr = 2 [36, 37] and the transition from the stable
solution given by Eq. (11) occurs only if the flow is perturbed on a lengthscale much larger than that of the base
flow [38]. In the absence of scale separation,
√ i.e. the ratio between the perturbation and the base flow wavenumber is
O(1), Re cr is slightly larger than 2. Accordingly, we fix Re = 100 and double the domain size to 2L = 4π, imposing
a low-wavenumber initial perturbation on k = 1/2. In this setting, the flow turns out to be unstable. We let the
flow evolve up to t = 500 and then add the fiber network. In particular, we have performed simulations considering
two different concentrations, N = 43 and 73 , with fiber length equal to c/L = (2π)−1 . In addition, to test the
validity of our simple model, we have conducted the same simulations replacing into the Navier Stokes equations the
fully-resolved IB approach for the fiber network with the effective Darcy’s term, i.e. fFIB = −Du, where D is given
by Eq. (10) (without adjusting the free parameter α in the expression).
The time history of the kinetic energy is reported in Fig. 7 for different fiber concentrations. The same phenomenol-
ogy already observed for the ABC flow can be recognized, with the overall stabilizing role of the network and, in
particular, a steady solution obtained for the highest concentration that has been tested. It is also important to note
that the behavior of the system forced only at the large scale by a friction Darcy-like term is in good agreement with
the results from the fully resolved simulations. The effective description is thus capable of capturing correctly the
dynamical state reached by the flow under the action of the fluid-structure coupling. Moreover, one can see that the
agreement improves for increasing concentrations; again, this is consistent with the fact that the nonlinear terms,
whose contribution to the spectral energy budget is shown in the inset of Fig. 7, decrease while increasing N , thus
making the validity of our model stronger.
Focusing on the steady case with higher fiber concentration, in Fig. 8 we compare the resulting velocity field,
similarly to what done previously for the cellular flows. While in the absence of fibers, for Re > Re cr , the stable
Kolmogorov flow solution (left panel) is lost to reach the unstable one (center panel), in the presence of a sufficiently
dense fiber network the fingerprint of the stable solution is recovered in the resulting large-scale configuration (right
panel). Finally, we compare the single-phase solution obtained at a Reynolds number equal to the effective Reynolds
number of the fiber-laden case (as previously done in Sec. III A), verifying that a similar steady solution is recovered.

C. Stationary homogeneous isotropic turbulence

As the final step of our analysis, we consider a widely used model for turbulent flows: the homogeneous isotropic
turbulence (HIT) [5]. To reproduce numerically this configuration, the flow is sustained using the spectral forcing
 10

0.022
N=0
0.02
N = 43 , IBM
0.018 N = 43 , eff.
N = 73 , IBM
0.016 5 × 10 −4
N = 73 , eff.
0.014
0 × 10 0
0.012
1 hu2 i

k T (k)
2

0.01 −5 × 10 −4

0.008
−1 × 10 −3
0.006

0.004 −2 × 10 −3
1 10
0.002
k
0
400 450 500 550 600

FIG. 7. Time history of mean fluid kinetic energy for different fiber concentrations in a Kolmogorov flow at Re = 100. Inset:
nonlinear terms appearing into the spectral balance Eq. (7) for the cases with fibers.

FIG. 8. Visualizations of one component of the velocity field for the Kolmogorov parallel flows. Note that the fluid domain
size is now 2L = 4π. Left: stable solution given by Eq. (11); center: unstable flow at Re = 100 (at a given time instant); right:
steady flow obtained in the presence of a network of N = 73 evenly spaced fibers with c/L = (2π)−1 .

scheme by Eswaran and Pope [42], where energy is injected randomly at low wavenumbers (in our case, within a
spherical shell with radius k = 2) by means of a Ornstein-Uhlenbeck process.
We choose a setting where, in the absence of fibers, the Reynolds number based on the Taylor’s microscale is
Re λ ≈ 40, and consider two different fiber concentrations, N = 43 and 103 , with c/L = (4π)−1 . As before, we have
simulated the flow without fibers up to a certain time t = 100, then used as the initial condition for the cases with fiber
network. Furthermore, we also carried out computations using the effective model for the same parameter setting,
based again on Eq. (10) for the Darcy’s coefficient.
First, we look at the kinetic energy in time, reported in Fig. 9. For the less concentrated case with N = 43 fibers,
the effect of the network appears to be very limited, while for N = 103 we have a significant reduction of the average
value, along with a decreased oscillation amplitude. Nevertheless, despite the stabilizing effect of the fiber network,
the flow always remains unsteady, clearly because of the random forcing that is applied. Comparing on a statistical
basis the fully-resolved cases with the corresponding effective ones, good agreement is found for both concentrations.
To check the consistency with the picture already drawn for the ABC and Kolmogorov flows, we look once again at
 11

0.4

0.35

0.3

0.25
2 hu i
1 2

0.2
N=0
0.15
N = 43 , IBM
0.1 N = 43 , eff.
N = 103 , IBM
0.05
N = 103 , eff.
0
0 20 40 60 80 100 120 140 160 180 200

FIG. 9. Time history of the mean fluid kinetic energy for different fiber concentrations in homogeneous isotropic turbulence at
Re λ ≈ 40.

(a) (b)
0.02 100

0.015 10−1
0.01
10−2
0.005 k−3
10−3
0
k T (k)

E(k)

10−4
−0.005
10−5
−0.01
10−6
−0.015
N=0 10−7 N=0
−0.02 N = 43 N = 43
N = 103 N = 103
−0.025 10−8
1 10 1 10
k k

FIG. 10. (a) Nonlinear term contribution of the spectral power balance according to Eq. (7) and (b) energy spectra in
homogeneous isotropic turbulence at Re λ ≈ 40 for different fiber concentrations (black: N = 0, red: N = 43 , blue: N = 103 )
with fiber length c/L = (4π)−1 .

the role of the nonlinear terms in the spectral balance, Eq. (7), shown in Fig. 10a: the same trend observed before for
the Kolmogorov flow (see the inset of Fig. 7) is found also here, with T (k) decreasing when increasing N .
In this regard, we point out that this scenario can change for a different choice of the governing parameters. For
example, we have conducted some tests (not shown here) where the viscosity is decreased and, in turns, T (k) becomes
more important. As a result, the effective description is found to underestimate appreciably the friction computed
with the fully-resolved approach. In such conditions, the need of a more elaborated model taking into account the
role of inertial terms is required.
Next, the energy spectra are shown in Fig. 10b. Overall, one can note the close resemblance with the trend observed
when considering the ABC cellular flow (Fig. 4a). On one hand, the large-scale/low-wavenumber components decrease
for increasing the fiber concentration, while the opposite occurs for the small-scales/high-wavenumbers, for which we
recover the same scaling ∼ k −3 , fingerprint of smooth-in-space velocity excursions. Furthermore, for the case with
the highest concentration we can clearly observe a region of low energy for intermediate wavenumbers and the peak
 12

at kc = 10, representing the signature of the non-local mechanism of energy distribution previously identified.

IV. CONCLUSIONS

We perform direct numerical simulation of three periodic flows with a network of fixed rigid fibers suspended within;
the presence of the fiber is simulated using an immersed boundary method which handles the fluid-structure coupling.
In particular, the Arnold-Beltrami-Childress cellular flow with closed streamlines, the parallel Kolmogorov flow with
open streamlines and the homogeneous isotropic turbulent flow are considered in order to understand the stability
and modifications of energy transfer of flows within canopies, fibrous media, and particle-laden flows.
First, we find that the fiber network has a stabilizing effect on the flows. Indeed, the ABC flow can be stationary
even at large Reynolds numbers in the presence of a high concentration of fibers, with the resulting stationary flow
mimicking the single phase stable solution at a lower Reynolds number. Based on this evidence, we therefore perform
separate analysis for the large and small scales of the flow. For the large-scale dynamics, we find that the effect of
increased drag exerted by the network of fibers on the flow can be effectively modelled by means of a Darcy’s friction
term. This can be used to model the large scale motion of the flow and is tested in all the flows considered here,
i.e. the ABC, Kolmogorov and turbulent flows. As concerns the small-scale dynamics, we find that the presence
of fibers triggers small-scale activity, which results in an energy spectrum with the emergence of a secondary peak
at a wavenumber corresponding to the the network spacing. By examining the overall energy distribution across
the various scales of motion, we find that the non-linear contribution to the energy balance rapidly vanishes as the
concentration of the network grows, with the fluid-structure coupling term balancing the external forcing at the
large-scales and the viscous dissipation at the small-scales. The fluid-structure coupling dissipates energy at the large
scale and re-introduces energy in the system at the small ones, thus effectively acting as a nonlocal energy transfer
mechanism.
This work highlights the key features of a fiber network on the flow. Our analysis clarified the origin of the
modifications of the energy spectrum in the presence of suspended rigid fibers, which however are common in several
other systems, ranging from canopy flows, flows in porous media and even suspension flows. At which extent our
findings apply when the fibers are freely moving into the flow is the subject of future investigations.

ACKNOWLEDGMENTS

SO acknowledges OIST for supporting his visiting period in the Complex Fluids and Flows Unit. AM thanks
the financial support from the Compagnia di San Paolo, project MINIERA n. I34I20000380007. LB acknowledges
financial support from the Swedish Research Council (VR), Grant No. VR 2014-5001. Computing time was provided
by INFN and CINECA.

Appendix A: Energy spectrum equation

In this appendix, we briefly recall how Eq. (7) is derived, along with identifying each term appearing in the equation.
A detailed explanation can be found in classical textbooks, see e.g. Pope [43].
As the starting point, we perform the Fourier transform of the Navier-Stokes Eqs. (1) and (2), yielding:

∂t û + Ĝ = −ikp̂/ρ0 − νk 2 û + f̂ , (A1)

k · û = 0, (A2)
ˆ
where (·)(k, t) = F {(·)(x, t)} denotes the Fourier transform, G corresponds to the nonlinear term appearing in the
momentum equation and k is the wavenumber vector. The same equations can be written for the complex conjugate
û∗ . Multiplying Eq. (A1) by û∗ , the pressure term drops due to the incompressibility constraint, Eq. (A2), and the
same applies in the momentum equation for û∗ when multiplying by û.
When summing the two equations for û and û∗ , we obtain an equation for the spectral kinetic energy Ê(k, t),
defined as Ê(k, t) = hû∗ · ûi/2, which reads as

∂t Ê = T̂ + V̂ + F̂ , (A3)

where we have identified the following quantities:

 13

1
– T̂ = 2 (Ĝ · û∗ + Ĝ∗ · û) is the transfer term associated with the nonlinear convective term;

– V̂ = −2νk 2 Ê is the viscous dissipation;

1
– F̂ = 2 (f̂ · û∗ + f̂ ∗ · û) is the energy input/output associated with the forcing.
Finally, to obtain Eq. (7) for the energy spectrum E(k, t), Eq. (A3) is averaged in each direction for isotropy, i.e. over
a sphere of radius k.

[1] B. Dubrulle and U. Frisch, Phys. Rev. A 43, 5355 (1991).

[2] M. Germano, U. Piomelli, P. Moin, and W. H. Cabot, Physics of Fluids A: Fluid Dynamics 3, 1760 (1991),
https://doi.org/10.1063/1.857955.
[3] S. Gama, M. Vergassola, and U. Frisch, Journal of Fluid Mechanics 260, 95126 (1994).
[4] U. Frisch, Lecture Notes, NCAR-GTP Summer School June 1987 , 219 (1987).
[5] U. Frisch, Turbulence: the legacy of A. N. Kolmogorov (Cambridge University Press, 1995).
[6] L. Biferale, A. Crisanti, M. Vergassola, and A. Vulpiani, Physics of Fluids 7, 2725 (1995),
https://doi.org/10.1063/1.868651.
[7] A. Mazzino, Phys. Rev. E 56, 5500 (1997).
[8] P. Castiglione, A. Crisanti, A. Mazzino, M. Vergassola, and A. Vulpiani,
Journal of Physics A: Mathematical and General 31, 7197 (1998).
[9] A. Mazzino, S. Musacchio, and A. Vulpiani, Phys. Rev. E 71, 011113 (2005).
[10] M. A. LeMone, W. M. Angevine, C. S. Bretherton, F. Chen, J. Dudhia, E. Fedorovich, K. B. Katsaros,
D. H. Lenschow, L. Mahrt, E. G. Patton, J. Sun, M. Tjernstrm, and J. Weil, Meteorol. Monogr. 59, 9.1 (2018),
https://doi.org/10.1175/AMSMONOGRAPHS-D-18-0013.1.
[11] M. R. Raupach and A. S. Thom, Annu. Rev. Fluid Mech. 13, 97 (1981), https://doi.org/10.1146/annurev.fl.13.010181.000525.
[12] D. Poggi, A. Porporato, L. Ridolfi, J. D. Albertson, and G. G. Katul, Boundary-Layer Meteorol. 111, 565 (2004).
[13] A. Di Bernardino, P. Monti, G. Leuzzi, and G. Querzoli, Boundary-Layer Meteorol. 165, 251 (2017).
[14] M. Ghisalberti and H. Nepf, Transport in Porous Media 78, 309 (2009).
[15] R. Shnapp, E. Shapira, D. Peri, Y. Bohbot-Raviv, E. Fattal, and A. Liberzon, Sci. Rep. 9, 7405 (2019).
[16] G. A. Zampogna and A. Bottaro, Journal of Fluid Mechanics 792, 535 (2016).
[17] G. A. Zampogna, F. Pluvinage, A. Kourta, and A. Bottaro, Water Resources Research 52, 5421 (2016),
https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1002/2016WR018915.
[18] A. Monti, M. Omidyeganeh, and A. Pinelli, Physics of Fluids 31, 065108 (2019), https://doi.org/10.1063/1.5095770.
[19] J. Finnigan, Annu. Rev. Fluid Mech. 32, 519 (2000), https://doi.org/10.1146/annurev.fluid.32.1.519.
[20] H. M. Nepf, Annual Review of Fluid Mechanics 44, 123 (2012), https://doi.org/10.1146/annurev-fluid-120710-101048.
[21] D. Galloway and U. Frisch, Journal of Fluid Mechanics 180, 557564 (1987).
[22] O. Podvigina and A. Pouquet, Physica D: Nonlinear Phenomena 75, 471 (1994).
[23] O. du Roure, A. Lindner, E. N. Nazockdast, and M. J. Shelley, Annu. Rev. Fluid Mech. 51, 539 (2019).
[24] M. E. Rosti, A. A. Banaei, L. Brandt, and A. Mazzino, Phys. Rev. Lett. 121, 044501 (2018).
[25] M. E. Rosti, S. Olivieri, A. A. Banaei, L. Brandt, and A. Mazzino, Meccanica (2019), 10.1007/s11012-019-00997-2.
[26] A. A. Banaei, M. E. Rosti, and L. Brandt, Journal of Fluid Mechanics 882, A5 (2020).
[27] M. Cavaiola, S. Olivieri, and A. Mazzino, arXiv preprint arXiv:1908.04072 (2019).
[28] W.-X. Huang, S. J. Shin, and H. J. Sung, J. Comput. Phys. 226, 2206 (2007).
[29] A. M. Roma, C. S. Peskin, and M. J. Berger, J. Comput. Phys. 153, 509 (1999).
[30] M. E. Rosti and L. Brandt, Journal of Fluid Mechanics 830, 708 (2017).
[31] T. Dombre, U. Frisch, J. M. Greene, M. Hénon, A. Mehr, and A. M. Soward, J. Fluid Mech. 167, 353391 (1986).
[32] V. I. Arnold, in Vladimir I. Arnold-Collected Works (Springer, 1965) pp. 15–18.
[33] X. Zhao, K. Kwek, J. Li, and K. Huang, SIAM Journal on Applied Mathematics 53, 71 (1993),
https://doi.org/10.1137/0153005.
[34] L. Biferale, A. Crisanti, M. Vergassola, and A. Vulpiani, Phys. Fluids 7, 2725 (1995).
[35] A. Bottaro, Journal of Fluid Mechanics 877, P1 (2019).
[36] L. Meshalkin and I. Sinai, Journal of Applied Mathematics and Mechanics 25, 1700 (1961).
[37] A. Thess, Physics of Fluids A: Fluid Dynamics 4, 1385 (1992), https://doi.org/10.1063/1.858415.
[38] G. Boffetta, A. Celani, A. Mazzino, A. Puliafito, and M. Vergassola, Journal of Fluid Mechanics 523, 161170 (2005).
[39] G. Boffetta, A. Celani, and A. Mazzino, Phys. Rev. E 71, 036307 (2005).
[40] A. Bistagnino, G. Boffetta, A. Celani, A. Mazzino, A. Puliafito, and M. Vergassola,
Journal of Fluid Mechanics 590, 6180 (2007).
[41] J. Tithof, B. Suri, R. K. Pallantla, R. O. Grigoriev, and M. F. Schatz, Journal of Fluid Mechanics 828, 837866 (2017).
[42] V. Eswaran and S. Pope, Computers & Fluids 16, 257 (1988).
[43] S. B. Pope, Turbulent Flows (Cambridge University Press, 2000).