You are on page 1of 15

Microfluidics and Nanofluidics (2018) 22:20

https://doi.org/10.1007/s10404-018-2041-9

RESEARCH PAPER

Inertia‑driven enhancement of mixing efficiency in microfluidic


cross‑junctions: a combined Eulerian/Lagrangian approach
Alessia Borgogna1 · Maria Anna Murmura1 · Maria Cristina Annesini1 · Massimiliano Giona1 · Stefano Cerbelli1 

Received: 15 October 2017 / Accepted: 18 January 2018 / Published online: 29 January 2018
© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract
Mixing of a diffusing species entrained in a three-dimensional microfluidic flow-focusing cross-junction is numerically
investigated at low Reynolds numbers, 1 ≤ Re ≤ 150 , for a value of the Schmidt number representative of a small solute
molecule in water, Sc = 103 . Accurate three-dimensional simulations of the steady-state incompressible Navier–Stokes equa-
tions confirm recent results reported in the literature highlighting the occurrence of different qualitative structures of the
flow geometry, whose range of existence depends on Re and on the ratio, R, between the volumetric flowrates of the imping-
ing currents. At low values of R and increasing Re, the flux tube enclosing the solute-rich stream undergoes a topological
transition, from the classical flow-focused structure to a multi-branched shape. We here show that this transition causes a
nonmonotonic behavior of mixing efficiency with Re at constant flow ratio. The increase in efficiency is the consequence
of a progressive compression of the cross-sectional diffusional lengthscale, which provides the mechanism sustaining the
transversal Fickian flux even when the Peclet number, Pe = Re Sc , characterizing mass transport, becomes higher due to
the increase in Re. The quantitative assessment of mixing efficiency at the considerably high values of the Peclet number
considered ( 103 ≤ Pe ≤ 1.5 × 105 ) is here made possible by a novel method of reconstruction of steady-state cross-sectional
concentration maps from velocity-weighted ensemble statistics of noisy trajectories, which does away with the severe numeri-
cal diffusion shortcomings associated with classical Eulerian approaches to mass transport in complex 3d flows.

Keywords  X-Junction · Mixing efficiency · Flux tube · Diffusion · Numerical diffusion · Langevin equation · Concentration
variance

1 Introduction

Microfluidics-assisted mixers have attracted ever growing


Electronic supplementary material  The online version of this attention in the last two decades in view of their potential to
article (https​://doi.org/10.1007/s1040​4-018-2041-9) contains achieve rapid contact between segregated liquid streams at
supplementary material, which is available to authorized users. lengthscales where flow and temperature conditions can be
accurately controlled (Hessel et al. 2005; Zhu and Ye 2010;
* Stefano Cerbelli
stefano.cerbelli@uniroma1.it Liu et al. 2012). Owing to the reduced lengthscales, these
devices are typically operated in the laminar fluid dynamic
Alessia Borgogna
alessia.borgogna@uniroma1.it regime, often under strictly creeping (Stokes) flow condi-
tions. In several applications (Stremler et al. 2004), geom-
Maria Anna Murmura
mariaanna.murmura@uniroma1.it etries and operating conditions of continuous inflow–outflow
micromixers have been inspired by the principles of chaotic
Maria Cristina Annesini
mariacristina.annesini@uniroma1.it advection, originally developed in the framework of tran-
sient processes occurring in closed impermeable systems
Massimiliano Giona
massimiliano.giona@uniroma1.it (Aref et al. 2014; Gubanov and Cortelezzi 2010). Microflu-
idic implementations of these principles have been proved to
1
Dipartimento di Ingegneria Chimica Materiali Ambiente, increase the efficiency of a number of clinical applications
Sapienza Università di Roma, Via Eudossiana 18, Rome,
Italy

13
Vol.:(0123456789)
20 
Page 2 of 15 Microfluidics and Nanofluidics (2018) 22:20

(see, e.g., Cortelezzi et al. 2017; Raynal et al. 2007, 2013; a liquid mixture, thus yielding Pe−1 ≤ 10−5 (2) the bound-
Leung et al. 2015). ary conditions for mass and vorticy transport are different
However efficient, laminar chaos is not always straight- in the vast majority of the boundary value problems that are
forwardly attainable in open flow devices, and sometimes meant to represent a physical experiment, and (3) in a three-
is not even desirable in that the repeated stretching and dimensional setting the full vector character of the vorticity
folding of the partially mixed structures typically generates gives rise to the additional source (stretching) term 𝝎 ⋅ ∇𝐯
highly nonuniform micromixing intensities throughout the (where 𝝎 and 𝐯 represent the vorticity and the velocity vec-
mixing domain (Cerbelli et al. 2002). This is the case, e.g., tors, respectively) in the transport equation. Thus, if some
of microfluidics-assisted flash nanoprecipitation of poly- analogies can be expected in the two-dimensional setting
mers by solvent displacement (Saad and Prudhomme 2016; between scalar and vorticity transport, important differences
Chronopoulou et al. 2014), where nonuniform mixing may arise in the three-dimensional case.
result in a wide size distribution of polymeric nanoparti- By these observations, one surmises that, in open micro-
cles. Here simplicity and flow uniformity is favored over mixers operating at steady state, the connection between
mixing efficiency, and elementary flow-focusing geometries flow structure and mixing efficiency should be not—in gen-
such as T-, Y-, and X-shaped junctions have been considered eral—directly inferred from the vorticity field, but rather
(Zhao et al. 2011). It has been found that narrow distribu- should be sought in the geometry of the flux tube associated
tions of particle size can be obtained in accurately tuned with the solute-rich stream entering the device. A step for-
flow regimes, where the average particle diameter can reach ward in this direction has been taken only very recently by
values as low as 20 nm (Chronopoulou et al. 2014). Another Damian et al. (2017), who investigated both numerically and
example of how the strict control over the flow features can experimentally the structure of such flux tube in a microflu-
yield practical advantages is provided by polymer reaction idic cross-junction for a variety of operating conditions. A
engineering. Here, an accurate design of the device geom- detailed investigation of the partially mixed structures pre-
etry not only permits the fine regulation of specific product dicted by the numerical simulations and confirmed by the
features (e.g., the molecular weight distribution), but also experiments unveiled the occurrence of topologically differ-
can orient and direct the reaction pathways so as to obtain a ent features of the solute-rich stream, each characterized by
prescribed chemical composition along the polymeric chain its own region of existence in the plane R−Re . The question
(Bally et al. 2011). that arises from this interesting physical phenomenology,
For these reasons, detailed numerical studies addressing ultimately associated with the increasing influence of flow
both the steady-state flow structure and its stability under inertia at increasing values of Re, concerns the impact of
small perturbations have been performed in different micro- the different fluid dynamics regimes on advective-diffusive
mixer geometries (see, e.g., Lashgari et al. 2014; Oliveira transport of the solute, and, more specifically, how the geom-
et al. 2012; Sousa et al. 2011; Chen et al. 2015) typically etry of the flux tube affects the degree of mixing achieved.
characterized by a rectangular cross section. In this article, we address this yet unanswered question
As regards the specific case of the T-junction geometry, through a theoretical/numerical approach in a fully three-
extensive research effort has been put forward by Mauri dimensional setting. In this regard, the first problem to be
and co-workers, who analyzed different phenomenological faced is that classical Eulerian approaches that accurately
aspects of transport, ranging from the fluid dynamic char- solve the flow field cannot be expected equally accurate in
acterization of the flow features (Andreussi et al. 2015)— describing mass transport. This is because the high values
including the effects of composition-dependent viscosity and of Pe (compared to Re) that characterize mass transport in
density (Orsi et al. 2013)— to the impact of the flow struc- liquid mixtures are typically associated with sharp boundary
ture on mixing of a diffusive species entrained in one of the layers of the concentration field, which require an exceed-
feeding streams (Galletti et al. 2015). In trying to establish ingly fine discretization of the domain to be solved accu-
a connection between flow features and mixing efficiency, rately. To overcome this intrinsic limitation of a full Eulerian
attention has been mainly focused on the structure of the solution of momentum and mass transport, we enforce a
vorticity field, often invoking the formal analogy between mixed Eulerian–Lagrangian approach, where a commercial
mass and vorticity transport equations. However, there are software (COMSOL Multiphysics) is used to solve for the
fundamental differences between the solutions to these equa- Navier–Stokes equations and the resulting flow field is used
tions when a real world experiment involving liquid mix- to define the deterministic motion component of a swarm
ing is being dealt with: (1) molecular (diffusive) transport of advecting-diffusing particles entering the device at the
of vorticity is controlled by the factor Re−1 , which is typi- inlet of the solute-rich stream. The steady-state concentra-
cally of order 10−2 or higher in microfluidic applications, tion field is computed based on the box-counted density of
whereas the coefficient controlling mass diffusion is given the first intersection of these trajectories at a generic cross
by Pe−1 = (Re Sc)−1 where Sc ≃ 103 or higher for a solute in

13
Microfluidics and Nanofluidics (2018) 22:20 Page 3 of 15  20

section of the channel downstream the impinging region. significantly larger than unity) on mass transport and mixing
We provide theoretical arguments to show that a weighting can be captured even in the modeling framework where the
factor based on the streamwise velocity component must be mixture properties are assumed independent of composition.
enforced to obtain the steady-state concentration cross-sec- Under this hypothesis the steady-state momentum and mass
tional profile from the box-counted density of intersections. transport equation for the dimensionless velocity 𝐯 and sol-
ute concentration c can be written in the one-way-coupling
form as
2 Statement of the problem
1 2
𝐯 ⋅ ∇𝐯 = ∇ 𝐯 − ∇P [∇ ⋅ 𝐯 = 0] (1)
Re
2.1 System geometry and dimensionless
formulation of the problem
1 2
𝐯 ⋅ ∇c = ∇ c (2)
The geometric structure and the main operating parameters Pe
of the cross-junction are sketched in Fig. 1. where Re = U𝓁∕𝜈 ( 𝜈 being the kinematic viscosity of the
The solute enters the device from the top cross section, mixture), and Pe = U𝓁∕ = Re Sc (  being the diffusiv-
entrained in a solvent current of flowrate Q. A pure sol- ity of the solute in the mixture). In this context, by one-
vent stream of flowrate Qw is split into two equal currents way-coupling we mean that Eq. (1) can be regarded as stand
symmetrically entering the cross-junction from the lateral alone, and that mass transport can be solved ex-post once
feeding sections. The inlet and outlet cross sections are all the flow field 𝐯(𝐱) has been computed. As regards momen-
squared and identical. The half-length of the square edge, tum transport, no-slip and impermeability conditions are
𝓁 , is taken as reference length. The right panel of the figure assumed on all the solid surfaces of the microfluidic channel,
shows the relevant dimensions of the geometry, defined with laminar inlet conditions associated with an entrance length
respect to 𝓁 . The quantity Q∕Qw is henceforth referred to lin = 10 (thus giving rise to a fully developed Poiseuille
as flow ratio R. Next, we choose the average flow velocity profile) are enforced on all the inlet sections, whereas con-
U in the discharge channel (i.e., for z̃ ≥ 1 ) as a character- stant discharge pressure is assumed at the outlet section. As
istic velocity, and the concentration Cr of the solute in the concerns the local mass balance, Eq. (2), unit concentration
feeding stream as reference concentration. In principle, the is enforced at the top cross section, where the solute-rich
solutions of mass and momentum transport equations are stream enters the device, vanishing concentration is set at
coupled because the density and viscosity of the mixture the lateral feeding cross sections, and Danckwerts’ boundary
depend on its composition, an occurrence that makes the condition is assumed at the outlet of the device. Also, zero
transport model strongly nonlinear even at low values of Re normal component of the concentration gradient is enforced
and restricts the domain of numerical and theoretical inves- on all of the solid boundaries.
tigation to a handful of simple prototypical cases (see, e.g,
Sadeghi 2018; Galletti et al. 2015) Besides, as discussed in
the remainder of this article, physically relevant effects of
the impact of flow inertia (i.e., quantified by values of Re

Fig. 1  Left panel: 3d geometry 2


and feeding condition. Right
panel: sketch of the device Q = R Qw
projection on the yz plane. All
lengths are made dimensionless
Qw /2 4
with respect to half of the edge (a)
length of the (squared) inlet and 30 0 2
outlet sections. The origin of y
the coordinate system is located z
at the center of the impinging
Qw /2
region
x

y
z
Qtot = (1 + R) Qw

13
20 
Page 4 of 15 Microfluidics and Nanofluidics (2018) 22:20

2.2 Assessment of mixing efficiency Because the total (dimensionless) volumetric flowrate Qtot
(see Fig. 1) is constant for 1 ≤ z̃ ≤ 25 due to flow incom-
The general setting for the assessment of mixing efficiency pressibility, one gathers that the cup-average concentration
in open inflow–outflow systems operating at steady-state
has been developed by some of these authors in a series of ∫S vz c dSz̃ ∫S vz cin dSin R
(8)
z̃ in
c(̃z) = = = c =c
previous articles (see, e.g., Cerbelli et al. 2008, 2009). Here, ∫S vz dSz̃ Qtot 1 + R in
we report only the main concepts that are useful to set up a

quantitative framework for interpreting the performance of is constant in the assigned z̃ interval. Note that in this rela-
the micromixer geometry that constitutes the focus object of tionship cin is the dimensionless uniform concentration value
this article. To this end, it is convenient to set up compact at the top entrance of the device, here assumed uniformly
notation for the domain where mixing takes place, hence- equal to unity. The scalar variance 𝜎(̃z) , providing a global
forth denoted with Ω , as well as for its boundary 𝜕Ω . With measure of the distance between the actual concentration
specific reference to the dimensionless coordinate represen- profile c(x, y, z̃) and the perfectly mixed condition c = c(̃z)
tation depicted in Fig. 1, let is defined as
Sin = {(x, y, z) | |x| ≤ 1, |y| ≤ 1; z = − 5}
1 ( )2
SI = {(x, y, z) ||x| ≤ 1, |z| ≤ 1; y = − 5} 𝜎 2 (̃z) = v c − c dSz̃ (9)
(3) Qtot ∫Sz̃ z
SII = {(x, y, z) | |x| ≤ 1, |z| ≤ 1; y = 5}
Sout = {(x, y, z) | |x| ≤ 1, |y| ≤ 1; z = 25} Note that in order to interpret 𝜎 2 (̃z) as a global distance from
the homogeneous conditions, this quantity should be non-
denote the top, lateral, and outlet cross sections, respectively. negative, a condition that is satisfied if the axial velocity
Also, let component vz (x, y, z̃) is nonnegative at all the points of the
{ } assigned cross section. In all of the results that are next pre-
Sz̃ = (x, y, z) || |x| ≤ 1, |y| ≤ 1; z = z̃ (4) sented, the minimum value of z̃ was chosen so as to ensure
be a generic cross section of the channel downstream the that this condition is always met. The relationship between
impinging volume, 1 ≤ z̃ ≤ 25 , and Ωz̃ the mixing volume 𝜎 2 (̃z) and the structure of the three-dimensional concen-
between the entrance sections and Sz̃ . The solid walls will tration field can be obtained from Eq. (5) with the same
henceforth be collectively referred to as Sw . Also, 𝐧 will approach used above for obtaining the cup-average concen-
denote the local outwardly-oriented unit vector normal to tration from Eq. (2)
any of the boundary surfaces. The characterization of mixing R
(
R
)
efficiency is grounded upon appropriate volume integrals of 𝜎 2 (̃z) = c2in 1−
1+R 1+R
Eq. (2), together with the equation 2 (10)
− ∇c ⋅ ∇c dΩz̃
( ) 1 2 Pe Qtot ∫Ωz̃
∇ ⋅ 𝐯c2 = ∇ ⋅ ∇c2 − ∇c ⋅ ∇c, (5)
Pe Pe
where c2in is the dimensionless concentration value charac-
which is obtained by multiplying Eq. (2) by c, after taking
terizing the solute-rich stream entering the device, which
into account the incompressible nature of the flow. Integrat-
is here equal to unity given that this value was chosen as
ing Eq. (2) over the volume enclosed between the entrance
reference concentration, and Qtot is the total dimensionless
cross sections Sin , SI , SII , and the generic cross section Sz̃ at
volumetric flowrate in the discharge channel, Qtot = 4 . The
z̃ one obtains
first term at the r.h.s. of Eq. (10) can be regarded as the ini-
∑ ( )
1 tial squared variance, say 𝜎02 , stemming from the prescribed
c 𝐯 − ∇c ⋅ 𝐧 dS𝛼 = 0 (6)
𝛼
∫ S𝛼 Pe feeding conditions at all of the entrance cross sections of the
device. As can be readily inferred from Eq. (10), the scalar
where 𝛼 stands for “in”, “I”, “II”, and “ z̃ ”. Note that the inte- variance can only decrease at increasing z̃ , since the volume
gral over the solid boundaries Sw vanishes identically since Ωz̃ over which the integral at the r.h.s. of Eq. (10) increases
both 𝐯 ⋅ 𝐧 and ∇c ⋅ 𝐧 are equal to zero. If the axial diffusion with z̃  , and the squared gradient is nonnegative. Another
contribution is neglected with respect to the convective com- important conclusion that can be made from this equation,
ponent, the surface integrals over SI and SII vanish identically is that in the limit where Pe goes to infinity 𝜎 2 (̃z) → 𝜎02 ,
( c = 0 at the lateral inlets), and Eq. (6) reduces to i.e., no mixing occurs in the kinematic limit of vanishing
solute diffusivity.
∫Sin
vz c dSin =
∫Sz̃
vz c dSz̃ (7)

13
Microfluidics and Nanofluidics (2018) 22:20 Page 5 of 15  20

From all of these observations it follows that a dimen- vector-valued Wiener process in the time √ interval dt , pos-
sionless normalized measure of mixing efficiency, say 𝜂z̃ , sessing zero mean and variance equal to dt . More specifi-
can be defined as cally, one considers the ensemble statistics performed over
a large number of particles, say Np—initially distributed
𝜎(̃z)
𝜂z̃ = 1 − (11) throughout the mixing domain so as to satisfy the initial
𝜎0
concentration field c(𝐱, t = 0) = c0 (𝐱) . The evolving con-
The overall device efficiency, henceforth denoted by 𝜂 , is centration field is computed as the box-counted particle
therefore given by number density performed over a finite-size discretiza-
tion of the transport domain in subdomains of size 𝛿 . In
𝜎out
𝜂 =1− , (12) the limit where Np → ∞ , and 𝛿 → 0 , the box-counted den-
𝜎0 sity of particles converges to the (normalized) concentra-
where 𝜎out is the scalar variance measured at the device out- tion field that is solution of the unsteady transport equation
let. In what follows we use these quantities to characterize 𝜕t c = −∇ ⋅ (𝐯c) + (1∕Pe)∇2 c . This approach has been, in
the mixing performance of the device under different operat- fact, widely exploited to determine the mixing efficiency of
ing conditions. the most diverse stirring protocols in laminar (Vikhansky
2006), as well as turbulent (Dentz and Barros 2015) flow
regimes, especially in two-dimensional representations. It
3 Steady‑state concentration maps has been also implemented with some degree of success in
from ensemble statistics of noisy the analysis of inflow–outflow devices that are constituted
Lagrangian trajectories by a unique channel possessing a single inlet and a single
outlet (Singh et al. 2008; Gorodetskyi et al. 2014a, b). How-
As briefly pointed out in the Introduction, a major issue in ever, as discussed below, the implementation of this method
predicting the performance of real world micromixers pro- to multiple-inlet multiple-outlet devices can lead to serious
cessing liquid streams is related to the high value of the inconsistencies, especially when the flowrate of the solute-
Peclet number of the solutes subject to transport. In cases rich current is smaller than that of the pure solvent.
where a full three-dimensional model is needed to capture Next, we set forth on extending the Langevin-stochastic
the relevant physical phenomenology of momentum and approach to determine the steady-state concentration maps
mass transport, the computational effort required for deter- at the cross sections at constant z̃ in the discharge channel
mining an accurate solution of the advection–diffusion equa- of the device, so as to determine the performance of the
tion may become prohibitive. When the discretization of the device through Eq. (12). The underlying idea of the method
mass transport equation(s) is insufficient to accurately rep- next described is to regard cross-sectional mixing along the
resent concentration gradients, numerical diffusion under- streamwise coordinate z as a two-dimensional transient prob-
mines the prediction of mixing efficiency so that the actual lem taking place in closed domain, with the downstream
Pe value of the computation, i.e., resulting from the super- coordinate z playing the role of time. However, a straight-
position of true and numerical diffusion, can be orders of forward transposition of the box-counting approach to the
magnitude smaller than that of the targeted physical system. analysis of devices with multiple-inlet/-outlet sections can
For this reason, alternative approaches are needed, and dif- lead to conceptual and practical errors, ultimately causing
ferent approximations have been applied to avoid the direct the violation of the basic properties of the concentration
Eulerian approach for computing the concentration field field, namely the min–max principle and the conservation
(see, e.g., Matsunaga et al. 2013). of mass. In this respect, it is useful to start with the analysis
In transient mixing problems taking place in closed of the purely kinematic (diffusionless) limit of the transport
domains, the possibility of tackling the problem from a equation. In this setting, the steady-state advection–diffu-
computationally different angle is provided by the Fok- sion equation reduces to ∇ ⋅ (𝐯c) = 𝐯 ⋅ ∇c = Dc∕Dt = 0 .
ker–Planck equivalence between the Eulerian probability Thus, the entrance concentration value must be conserved
balance quantifying the evolving concentration field and the along a streamline, regardless of the computational method
stochastic process (Lasota and Mackey 2013) employed. This observation provides a guidance to set up
√ the method for reconstructing the concentration field from
dxi (t) = vi (𝐱(t)) dt + 2∕Pe d𝜉i (t) {i = 1, 2, 3}, (13) cross-sectional box counting of the intersections of particles
where dxi (t) is the i-th component of the total displacement trajectories, here coincident with the flow streamlines.
over the time dt of an advecting-diffusing particle subject to Consider a situation like that depicted in Fig. 2a, where
the deterministic action of the velocity field 𝐯 = (v1 , v2 , v3 ) , a two-dimensional setting has been assumed for simplic-
and to the action of Brownian fluctuations expressed by ity. Assume a unit concentration is fixed at the entrance
d𝜉i (t) , which is the i-th component of the increment of a (top) cross section, here represented by a one-dimensional

13
20 
Page 6 of 15 Microfluidics and Nanofluidics (2018) 22:20

∆S Ph onto the cross section at z = z̃  , where Ph is chosen so


that all of the streamlines entering the system through ΔS
∆lj = vz (Pj ) ∆t
at Pj cross the surface element ΔS at Ph . At steady-state, the
number Npart of particles that originally occupied the vol-
ume ΔS vz (Pj ) Δt are now distributed throughout a volume
ΔS vz (Ph ) Δt . Therefore, the contribution of the cross-sec-
tional element at the point Pj of the inlet section to the local
s (volumetric) concentration at the point Ph at z̃ is given by
Pj z
| vz (Pj )
Fh(̃z) || = Fj(in) , (15)
|Fj
(in) vz (Ph )
∆lh = vz (Ph ) ∆t
where “ ||F(in) ” reads “conditioned to”. Clearly, other stream-
j
Ph
∆S lines started at contiguous locations with respect to the sur-
face element ΔS at Pj also contribute to the overall value of
(a)
Fh(̃z) . Thus, for a generic time-independent discrete represen-
tation of inlet concentration profile, cin (Pj ) , where Pj belongs
to the inlet cross section, the overall contribution to the box-
counted approximation to the concentration cz̃ (Ph ) at a cross
section located at z = z̃ is given by

∑ vz (Pj )
cz̃ (Ph ) = cin (Pj ) Fj(in) , (16)
j∈𝕁
vz (Ph )

(b) where 𝕁 represents the set of all of the cells belonging to the
inlet section whose streamlines possess nonempty intersec-
tion with the surface element ΔS centered at the point Ph
Fig. 2  Pictorial representation of the mass balance constraint giving
rise to the velocity-weighted box-counting algorithm belonging to the cross section at z̃ . From this simple argu-
ment one gathers that in the purely convective limit each
intersection of a particle trajectory with a prescribed cross
interval. The flux tube of the solute-rich stream is shrunk section contributes to the local concentration value through
into a central lamella by lateral currents of fresh solvent (not the weighting factor vz (P0 )∕vz (Pz̃ ) , where P0 and Pz̃ are the
shown for simplicity). Note that the profiles of the vertical intersections of the trajectory with the inlet cross section
velocity component are nonuniform along both the entrance and the cross section at z̃ . One notes that in the limit where
and exit cross sections of the control volume. Let us fix a ΔS, Δt → 0 , the coarse-grained concentration field so
discretization of the cross sections into equal intervals of defined ensures mass conservation as well as conservation
size ΔS . Assume Np is the total number of streamlines, each of the concentration value when moving along a streamline,
associated with a particle entering the flow domain, uni- i.e., it satisfies all of the fundamental properties of the solu-
formly spaced onto the top cross section. Let Np be chosen so tion to the pure advection equation.
that for a generic interval of width ΔS , centered at the point Two arguments may seem to undermine the extension of
Pj , the box-counted density of intersections the above method to the case in which Brownian diffusion
is also present (see Fig. 2b), namely
Nj
Fj(in) = (14) 1. The presence of diffusion, however small, destroys any
Np ΔS
kinematic invariant associated with the flow, thus mak-
ing the physical argument used to derive Eq. (15) not
is uniformly equal to unity. In Eq. (14), Nj is the number
altogether straightforward.
of streamlines crossing the interval ΔS centered at Pj . The
2. The lateral surfaces of the volume elements depicted in
number of particles, say Npart , entering the control volume
Fig. 2a are permeable to the diffusive flow; therefore, the
through ΔS at Pj in the interval Δt is proportional through
particle number balance motivating the introduction of
the volumetric concentration to the local normal veloc-
the weighting factor needs no longer hold true.
ity vz (Pj ) . Consider the same area element ΔS at a point

13
Microfluidics and Nanofluidics (2018) 22:20 Page 7 of 15  20

The latter issue can be readily shown to be immaterial, in


that a choice of the time interval Δt << ΔS∕vz (P) in the limit
ΔS → 0 , Δt → 0 makes the lateral surface of the volume
element much smaller than that of the base surface element,
and therefore makes the mass flowrate through the lateral
surface negligible with respect to that associated with the
surface element ΔS . As regards the first issue, one notes that
even though the noiseless physical picture giving rise to the
relationship in Eq. (15) can be used as a conceptual frame-
work to introduce the box-counted concentration mapping
expressed by Eq. (16), the only true physical assumptions
for its validity are that (1) time-independent conditions char-
acterize the concentration field at the feeding cross section,
(2) steady-state conditions have been reached, and (3) con-
vective transport is the only mechanism by which the solute
enters and leaves the control volume at the feeding section
and the generic cross section at z̃ , respectively. Clearly, when
tracking the noisy trajectories, an integration time large
enough to guarantee that all of the particles starting at t = 0
at the domain inlet do actually cross the prescribed cross
section at z̃ must be allowed.
Based on these observations, we next use Eq. (16) to
determine the concentration fields stemming from different Fig. 3  A detail of the nonuniform mesh used for the numerical solu-
operating conditions at fixed Schmidt number and variable tion of the velocity field. A finer discretization has been enforced in
Re and R values. The method is first validated by comparison the region where the feeding currents impinge onto each other, where
steep velocity gradients are expected
with the Eulerian solution of the concentration field obtained
by using the COMSOL Multiphysics platform at values of
Pe ≤ 103 , that are expected low enough so as to yield a solu- explain why the symmetrical solution, which proves robust
tion to the mass transport problem that is not overshadowed with respect to small perturbations, can be actually observed
by numerical diffusion. in physical experiments. Let us next move to the analysis of
the kinematic structures stemming from different operating
conditions specified by the dimensionless parameters Re and
4 Results R. Figure 4 shows the structure of half of the flux tube of the
stream entraining the solute at low flow ratio, R = 5 × 10−3 ,
4.1 Kinematic structure of the flux tube which corresponds to an average velocity in the top feed-
ing channel a hundred times smaller than those of the pure
The steady-state momentum transport problem described in solvent streams entering the device from the lateral inlets.
Sect. 2 was solved using COMSOL Multiphysics commer- As can be observed, the simple lamellar structure of the
cial suite. Since steep gradients are expected in the region flux tube that is evident at Re = 1 (Fig. 4a) is progressively
where the feeding currents impinge onto each other, a finer destroyed as the Reynolds number increases. This topologi-
discretization is used to represent this region (see Fig. 3). cal transition of the kinematic flux tube is altogether general
Overall, the mesh consists of order 1.5 × 106 tetrahedral in microfluidic systems, and can be ascribed to the inter-
elements. action of flow inertia with sharp boundaries at the chan-
At values of Re such that the flow is steady, a twofold nel intersections (Damian et al. 2017; Wong et al. 2004).
symmetry about the xz and the yz planes passing through Recirculation loops in the impinging region are already
the origin of the coordinate system can be assumed, so in clearly appreciable at Re = 30 [panel (b) of the figure], until
principle one would only need to simulate one fourth of the a topologically rich structure develops for the flux tube at
structure. Our motivation for representing the entire domain Re = 100 , where the branches originated by the current
is that this choice, combined with the use of an unstruc- impact are increasingly attracted toward the high-velocity
tured—and therefore not symmetrical—mesh, indirectly pro- core of the channel when moving downstream the chan-
vides a robustness check on the symmetrical solution. Thus, nel intersection. It can be seen that the topological transi-
the results presented here also provide a further confirmation tion occurs via the formation of vortices at the center of
of those reported in Damian et al. (2017) and ultimately the impinging zone, whose axes are initially collinear with

13
20 
Page 8 of 15 Microfluidics and Nanofluidics (2018) 22:20

Fig. 4  Three-dimensional
kinematic structure of half of
the flux tube of the solute-rich
stream at fixed flowrate ratio
R = 5 × 10−3 . a–c Depict the
cases Re = 1 , Re = 30 , and
Re = 100 , respectively

the directions of the feeding streams and are then progres- area of the tube cross section is significantly reduced and its
sively bent until they become aligned to the z direction in the thickness shrunk to very small lengthscales. This difference
discharge channel. Note that this vortex structure is inher- between the average thickness of the flux tube, depending on
ently three-dimensional and, as such, cannot be captured by whether it becomes localized in the core or at the near-wall
a two-dimensional representation of the momentum trans- region of the discharge channel, ultimately driven by the
port equation. We observe that the qualitative transition of existence of a nonuniform axial velocity profile, is the fun-
the kinematic flux tube with increasing Re is a phenomenon damental mechanism explaining the nontrivial response of
that is qualitatively robust even with respect to changes of mixing efficiency at increasing values of the Reynolds num-
the device geometry not shown for brevity. ber that constitutes the main focus of the next section. We
A clearer representation of this phenomenon is reported also observe that the kinematic transition of the flux tubes
in Fig.   5, which shows the intersection of the solute- just described could not be captured by any two-dimensional
rich stream with planes at constant z̃  . The three columns representation of the system (Damian et al. 2017).
labeled a, b, and c represent different fluid dynamic con- To obtain the representation of Fig.  5, velocity data
ditions, namely Re = 100 and R = 5 × 10−3 , Re = 150 and obtained through the COMSOL Multiphysics software were
R = 5 × 10−2 , Re = 100 and R = 5 × 10−1 , respectively. The exported on a regular grid of order 7 × 106 nodes. The nodal
rows identify different cross sections, three located at the values were thus trilinearly interpolated so as to define a
top ( z̃ = − 1 ), the center ( z̃ = 0 ), and the bottom ( z̃ = + 1 ) locally smooth, globally continuous velocity field 𝐯int (𝐱) .
of the central cubic cell constituting the core of the imping- Starting from this vector field, trajectories originating from
ing region, two located at an intermediate position down- order 105 points uniformly distributed over the top feeding
stream the discharge channel and close to the device exit. We cross section at z = − 5 were computed as solutions of the
observe that in all of the cases shown, the structure of the ODE system d𝐱∕dt = 𝐯int (𝐱) by using a standard fourth-order
flux tube becomes z-invariant before reaching the channel Runge–Kutta integration scheme. Because the geometric
exit (see animated videos in the Supplementary Material), structure of the trajectories is independent of the magnitude
thus implying that the flow at the channel exit has converged |𝐯 (𝐱)| of 𝐯 (𝐱) (it only depends on its local direction), a
| int | int
to a Poiseuille profile for the values of Re considered. normalized velocity was enforced in the integration so as
For the operating conditions depicted, the Reynolds num- to speed up the computation. Results of this procedure are
ber is large enough so that the classical hydrodynamic focus- reported in Fig. 5 for different values of Re and R at different
ing structure cannot be observed, with the exception of case locations z̃ of the cross section
c at the exit of the impinging region ( z̃ = 1 ). However, this
single lamellar structure quickly develops into a double-T- 4.2 Steady‑state concentration field and mixing
shaped flux tube which is progressively pushed toward the efficiency
channel walls, there attaining a complex folded structure. At
low and intermediate values of the flow ratio (cases a and b) Cross-sectional concentration profiles were reconstructed
the flux tube develops into a convoluted structure possessing through the box-counting procedure expressed by Eq. (16).
a twofold symmetry that runs in the central core of the chan- We observe that in the purely convective limit this proce-
nel. Since the local velocity is high in this region, the overall dure is exact, meaning that when ΔS → 0 and the number

13
Microfluidics and Nanofluidics (2018) 22:20 Page 9 of 15  20

(a) (b) (c) the solution through the finite-element discretized approach
is expected to yield an accurate solution. In the present case,
a preliminary analysis (not discussed here for brevity) shows
z̃ = −1 that the limit Peclet value at which the solution obtained
through the package transport of dilute species of COMSOL
Multiphysics can be considered accurate is Pe ≃ 103 , when
the advection–diffusion equation is solved onto the same
computational mesh used for solving the Navier–Stokes
equation. Thus, we take this limit value to quantitatively
z̃ = 0 validate the Lagrangian approach. Figure 6 shows the con-
centration profiles along the center line x = 0 , − 1 ≤ y ≤ 1 at
the exit cross section of the device for two values of the flow
ratio R = 5 × 10−1 ; 5 × 10−2 at Re = 1 and Re = 10 [Panel
(a) and (b), respectively].
Because the Schmidt number was fixed at the ref-
z̃ = 1 erence value Sc = 103 , the first case corresponds to
Pe = Re Sc = 103 , the second to Pe = 104 . As can be
observed, the profiles stemming from the two different
approaches closely agree at Pe = 103 , whereas a significant
discrepancy is already evident at Pe = 104 , thus implying
that the computational grid used for the Eulerian solution
z̃ = 8 introduces significant numerical diffusion effects, which
are expected to grow unchecked when operating conditions
involving Re = 102 or higher (and therefore Pe ≃ 105 ) are
considered.
This much established, we next analyze the features of
the concentration field at those Re values where the Eule-
z̃ = 24 rian approach is unfeasible unless extremely large computing
facilities are affordable. Figure 7 depicts the concentration
fields of the diffusing species computed through the Lagran-
gian velocity-weighted box-counted average for different
y operating conditions.
Left and right columns in the picture represent the cross
sections at z̃ = 3 (i.e., just below the impinging region), and
x at z̃ = 24 , i.e., near the device outlet. The three rows corre-
spond to different fluid dynamic conditions, namely Re = 102
Fig. 5  Intersections of the flux tube of the solute-rich stream and R = 5 × 10−3  , Re = 1.5 × 102 and R = 5 × 10−2  ,
with planes at constant z̃ (see main text for details). a Re = 100 , Re = 102 and R = 5 × 10−1 , from bottom to top. In all of
R = 5 × 10−3 ; b Re = 150 , R = 5 × 10−2 ; c Re = 100 , R = 5 × 10−1 .
The black central square depicts the domain |x| ≤ 1; |y| ≤ 1 . Note the cases, because of the large Pe values, the patterns of the
that for − 1 ≤ z̃ ≤ 1 , the flux tube invades part of the lateral feeding concentration field closely mimic those of the correspond-
channels along the y direction (gray-shaded area) ing kinematic flux tube [compare, e.g., panels (f) and (g) of
Fig. 7 with the cases “a” and “c” at z̃ = 24 of Fig. 5]. How-
ever, the level of mixedness reached at the exit cross section
of trajectories goes to infinity, the factor based on the is rather different, depending on the operating conditions
velocity ratio is the only weighting function ensuring mass (note that the color scale of concentration values is not the
conservation within an arbitrary flux tube. Based on the same for all cases shown). From the analysis of the cases
physical arguments developed in Sect. 3, this relationship shown, one would be driven to conclude that the degree of
is expected to hold true even in the presence of spanwise homogenization of the diffusing species, and therefore the
diffusion. Thus, the most severe test to check the validity of efficiency of the microfluidic device, depends essentially on
the weighted box-counting approximation of the steady-state the flow ratio, and specifically, the lower the value of R, the
concentration field is provided by the cases characterized by higher the degree of mixedness. However, this conclusion is
(relatively) low values of the Peclet number. Besides, these readily disproved by the comparison of results obtained with
cases are precisely those for which a Eulerian approach to the same flow ratio, R, but different Re values. Consider,

13
20 
Page 10 of 15 Microfluidics and Nanofluidics (2018) 22:20

Fig. 6  Comparison between the 1


Eulerian (symbols) and Lagran- a 1
gian (continuous lines) solution a
of the steady-state concentration 0.75
field along the symmetry line 0.75
x = 0 , − 1 ≤ y ≤ 1 at Re = 1
( Pe = 103 ) (a), and Re = 10 0.5

c
0.5
( Pe = 104 ) (b). Labels “a” and
“b” refer to R = 5 × 10−1 and 0.25 b
R = 5 × 10−2 , respectively b 0.25

0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
y y
(a) (b)

Fig. 7  Concentration maps
computed from velocity- z̃ = 3 z̃ = 24
weighted box-counted intersec- 1 1 1 1
tions of noisy trajectories with
the planes at z̃ = 3 (left column) 0.8 0.8
0.5 0.5
and z̃ = 24 (right column). Let-
tering is consistent with Fig. 5. 0.6 0.6
a Re = 102 , R = 5 × 10−3 b 0 0 (c)
y

y
Re = 1.5 × 102 , R = 5 × 10−2 ; 0.4 0.4
c Re = 102 , R = 5 × 10−1 . Note -0.5 -0.5
that the range of the color scale 0.2 0.2
is different depending on the
case (color figure online) -1 0 -1 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x x

1 1 1 1

0.8 0.8
0.5 0.5
0.6 0.6
0 0 (b)
y

0.4 0.4
-0.5 -0.5
0.2 0.2

-1 0 -1 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x x

1 0.25 1 0.25

0.2 0.2
0.5 0.5
0.15 0.15
0 0 (a)
y

0.1 0.1
-0.5 -0.5
0.05 0.05

-1 0 -1 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x x

13
Microfluidics and Nanofluidics (2018) 22:20 Page 11 of 15  20

Fig. 8  Comparison between the


kinematic structure of the flux
tube of the solute-rich stream
and the concentration profiles
for Re = 80 ; R = 5 × 10−2 .
The left column represents the
kinematic structures, the right
column the concentration maps.
a, b: z̃ = 3 ; c, d z̃ = 13 ; e, f
z̃ = 24 (color figure online)

for instance, the case Re = 80 , R = 5 × 10−2 represented in near-wall region the velocity is low, flow incompressibility
Fig. 8, where the concentration maps at different cross sec- dictates that the thickness of the tube must retain relatively
tions of the discharge channel are reported together with the large values. The corresponding concentration maps show
corresponding intersection of the kinematic flux tube. that cross-sectional mixing proceeds slowly in these condi-
At this combination of flow ratio and Reynolds number tions when moving downstream the device, as can be gath-
values, the kinematic flux tube entraining the solute-rich ered by the fact that concentration values in the core region
stream is pushed toward the channel walls. Since in the of the flux tube at the channel outlet are still very close to

13
20 
Page 12 of 15 Microfluidics and Nanofluidics (2018) 22:20

unity. The situation is rather different when increasing flow The quantitative representation of this conclusion is pro-
inertia causes the flux tube to be pushed away from the wall vided by the normalized scalar variance and of the overall
toward the core of the discharge channel, as was observed mixing efficiency, defined by Eqs. (9) and (12), respectively.
at the same flow ratio R = 5 × 10−2 but for a slightly higher Since the values of the above quantities may be affected
Reynolds value, Re = 1.5 × 102 (see the central column of by the degree of discretization of the domain over which
Fig. 5 and case b of Fig. 7). Here, because of the large veloc- they are computed, we performed a preliminary sensitiv-
ity values, the flux tube is swiftly squeezed to a thin convo- ity analysis to determine the impact of this parameter (not
luted shape. As a consequence, cross-sectional concentration shown for brevity). The results showed that the dependence
gradients are amplified by the cross-sectional components of of the curve on the box size becomes essentially immaterial
the convective flow, and mixing proceeds faster with respect when discretizations of order 200 × 200 boxes or finer are
to the lower Re case at the same flow ratio. considered. Based on these results, we fixed the size of the
Thus, the main qualitative conclusion that can be drawn discretization to a 300 × 300 grid and computed the over-
from the analysis of the structure of the concentration fields all mixing efficiency for three representative values of the
is that the influence of fluid dynamics on cross-sectional flow ratio, R = 5 × 10−3 ; 5 × 10−2 ; 5 × 10−1 , and Reynolds
mixing of the diffusing solute occurs through the sensitive numbers ranging from Re = 1 up to Re = 1.5 × 102 . Before
dependence of the geometry of flux tube entraining the sol- commenting these results, it is worth pointing out that if flow
ute-rich stream. High mixing efficiency is expected when inertia had no impact on the kinematic structures of the flow,
the combination of flow inertia and flow ratio is such as to a monotonically decreasing trend of the overall efficiency
yield a flux tube that becomes progressively localized in the 𝜂 versus Re should be expected, in that, at fixed Sc, the Pe
central (core) region of the discharge channel, since the local number increases proportionally with Re.
high-velocity values imply—by flow incompressibility—low In order to single out the effect of the Pe number from the
average thickness of the kinematic flux tube. inertia-driven change of geometry of the flow kinematics,

A B

C
1

0.8

0.6
η

D
0.4

0.2

0
103 104 105
Pe
Fig. 9  Mixing efficiency at increasing Pe for a flow ratio R = 5 × 10−2 . Empty symbols: fixed Sc = 103 and increasing Re; Solid symbols: fixed
Re = 1 and increasing Sc. The insets depict the concentration fields at z̃ = 24

13
Microfluidics and Nanofluidics (2018) 22:20 Page 13 of 15  20

in Fig. 9 we compared the performance of different pro- of changing geometry of the kinematic flux tube entraining
tocols as a function of the Pe number for fixed flow ratio the solute-rich stream, which, at fixed flow ratio, is driven
R = 5 × 10−2. by the increasing impact of flow inertia.
The empty symbols ( Δ ) report the values of efficiency at A synoptic representation of the mixing response of the
Sc = 103 and increasing Re. Here a minimum of efficiency microfluidic device versus Re at different values of the flow
can be observed at Pe = 8 × 104 (corresponding to Re = 80 ), ratio R is shown in Fig. 10. A sensitivity analysis of the data
followed by a rapid efficiency increase up to Pe = 1.5 × 105 . reported in the figure to the numerical details of the Lagran-
The filled symbols ( ▴ ) are instead obtained by holding the gian approach was carried out by sub-dividing the ensemble
Reynolds number fixed at Re = 1 , and increasing Sc. Thus, of noisy trajectories into ten equally sized, nonoverlapping
in the last case, the kinematic features of the flow do not subsets of initial particle positions. In all of the cases shown
change as Pe is increased, and the geometry of the flux tube the relative error of efficiency resulted of order 1% or below.
consists of a hydrodynamically focused central lamellar The values of R increase by two decades, from
structure, which is qualitatively analogous to the structure R = 5 × 10−3 of the upper curve to R = 5 × 10−1 of the lower
depicted in Fig. 4a. The comparison of the two curves shows curve. For the lowest flow ratio (upper curve) a complex
that the effect of increasing the Reynolds number is first to trend exhibiting both a minimum and a maximum efficiency
decrease efficiency, even below the value that pertains to the is found. By the observations put forward above, this behav-
lamellar-shaped flux tube associated with the creeping flow ior can be readily interpreted in terms of flow kinematics.
regime, until a threshold value is reached, beyond which the The range of Re associated with increasing efficiency are
trend is abruptly inverted. It is interesting to note that low those for which the kinematic flux tube entraining the sol-
efficiency in the inertial flow occurs whenever the solute- vent is progressively pushed away from the walls to become
rich stream is localized at the channel walls (the insets in the localized in the channel core. After the localization has
figure show the concentration contours at the device exit). occurred, efficiency starts decreasing again because of the
Conversely, after the topological transition of flow geometry increasing Pe values, while the kinematic structures of the
has occurred, the solute-rich stream is squeezed into such flow remain qualitatively unaltered. As regards the response
a thin structure that the amplification of transversal con- at relatively high values of R ( R = 5 × 10−1 , lower curve of
centration gradients can overshadow even the effect of the the figure), a complete detachment of the solute entraining
increasing values of Pe. tube from the device walls is never observed in the entire
From this preliminary observation it follows that the range of Re values considered, and the slowing of the effi-
existence of extremal points (be them minimal or maximal ciency decrement at increasing Re can be ascribed to the fact
extremals) in the efficiency response at fixed flow ratio R that the lamellar structure formed at the impinging region
and increasing Re are to be entirely ascribed to the impact still undergoes some degree of shrinking as it is progres-
sively transformed into the four-lobed geometry approaching
all of the channel walls (see Fig. 5).
As can be gathered by the values depicted in Fig. 10, the
1
incremental effect of flow inertia on mixing efficiency—
which was qualitatively highlighted when describing the
0.8 structure of the concentration fields—can be quantitatively
quite substantial. For instance, the analysis of the upper
curve of the figure (corresponding to the lowest flow ratio,
0.6 R = 5 × 10−3 ) shows that the same efficiency is obtained at
Re = 102 and Re = 15 . An analogous observation holds true
η

for the intermediate curve of Fig. 10a, where the efficiency


0.4
reached at Re = 1.4 × 102 is equal to that associated with a
value of the Reynolds number almost ten times smaller. On
0.2 a practical level, this implies that the same level of efficiency
is obtained when, for a fixed device geometry, temperature,
and nature of the solute, the overall (dimensional) flowrate
0 is increased by almost a decade. Thus, by accurately tuning
1 10 100
Re the fluid dynamic conditions so that the geometry of the kin-
ematic flux tube entraining the solute-rich current possesses
favorable features as regards the amplification of transversal
Fig. 10  Mixing efficiency at increasing Re and fixed flow ratio R.
The Schmidt number is set to Sc = 103 in all of cases shown. (∇) : concentration gradients, productivity can be increased size-
R = 5 × 10−3 ; (Δ) : R = 5 × 10−2 (◦) : R = 5 × 10−1 ably for one and the same physical system.

13
20 
Page 14 of 15 Microfluidics and Nanofluidics (2018) 22:20

5 Conclusions Villermaux (2012) for predicting the rate of homogenization


from local stretching rates in two-dimensional settings.
The mixing efficiency of a solute entering a flow-focusing A different—yet practically relevant—approach would
microfluidic cross-junction with squared cross section is be to perform a study where target fluid dynamic operating
numerically investigated in a three-dimensional setting at conditions are chosen, and mixing efficiency is investigated
increasing values of the Reynolds number, ranging from for Sc numbers varying in the range 104 ≃ 105 , representa-
creeping flow conditions to weakly inertial regimes, and for tive of relatively large solute molecules (e.g., of biological
flow ratios spanning two orders of magnitude. The analysis or clinical interest). Since mixing efficiency can be expected
presented, motivated by recent results highlighting nontrivial to decrease significantly when Sc is increased to such values
effects of flow inertia on the kinematic features induced by at constant R and Re, device geometries with remarkably
the velocity field (Damian et al. 2017), is here made pos- longer discharge channels should be considered in this anal-
sible by a novel method for computing the steady-state ysis. In this respect, the fact that at relatively low Re (i.e., in
concentration field onto prescribed cross- sections, which the range analyzed in this article) the velocity fields swiftly
is based upon velocity-weighted box-counted averages for converges to the Poiseuille (z-invariant) profile downstream
an ensemble of particle trajectories undergoing a stochastic the impinging region allows to devise a tailored computa-
advecting-diffusing process. This method proves an efficient tional approach, where the method described here is used
alternative to the direct Eulerian approach to the solution of in the first part of the geometry, and a spectral pseudo-2d
the advection–diffusion equation in that it does away with approach is enforced to analyze mixing in the discharge
numerical diffusion issues that typically undermine localized channel far from the impinging region.
discretization approaches to solute transport when values
of the Peclet number of order 103 or above are considered. References
The analysis of the partially mixed structures exhibited
by the diffusing species downstream the microfluidic device Andreussi T, Galletti C, Mauri R, Camarri S, Salvetti MV (2015) Flow
in different fluid dynamic conditions highlights a rich and regimes in t-shaped micro-mixers. Comput Chem Eng 76:150–159
diversified phenomenology, where different degrees of mix- Aref H, Blake JR, Budišić M, Cartwright JH, Clercx HJ, Feudel U,
Golestanian R, Gouillart E, Guer YL, van Heijst GF et al (2014)
edness are obtained depending on the geometric structure Frontiers of chaotic advection. arXiv preprint arXiv​:1403.2953
of the kinematic flux tubes entraining the solute-rich stream Bally F, Serra CA, Hessel V, Hadziioannou G (2011) Micro-
entering the device. The qualitative conclusion of this analy- mixer-assisted polymerization processes. Chem Eng Sci
sis is that mixing efficiency is increased whenever the com- 66(7):1449–1462
Cerbelli S, Alvarez M, Muzzio F (2002) Prediction and quantifi-
bination of flow ratio and flow inertia (quantified by Re) cation of micromixing intensities in laminar flows. AIChE J
gives rise to a kinematic flux tube that becomes localized 48(4):686–700
in the central core of the discharge channel. Under these Cerbelli S, Garofalo F, Giona M (2008) Steady-state performance of
conditions, the transversal lengthscales of the flux tube can an infinitely fast reaction in a three-dimensional open stokes flow.
Chem Eng Sci 63(17):4396–4411
be shrunk by orders of magnitude, thus amplifying cross- Cerbelli S, Giona M, Garofalo F, Adrover A (2009) Characterizing
sectional concentration gradients and sustaining the trans- relaxation timescales and overall steady-state efficiency of contin-
versal Fickian flux. This observation ultimately explains how uous inflow–outflow micromixers. La Houille Blanche 6:135–142
conditions that correspond to increasing values of the Pe Chen KK, Rowley CW, Stone HA (2015) Vortex dynamics in a pipe
t-junction: recirculation and sensitivity. Phys Fluids 27(3):034107
number can result in an increase in mixing efficiency. Con- Chronopoulou L, Sparago C, Palocci C (2014) A modular microfluidic
versely, when the flux tube entraining the solute is pushed platform for the synthesis of biopolymeric nanoparticles entrap-
toward the channel walls, transversal lengthscales are not ping organic actives. J Nanopart Res 16(11):2703
reduced so efficiently because of the locally low values of Cortelezzi L, Ferrari S, Dubini G (2017) A scalable active micro-mixer
for biomedical applications. Microfluid Nanofluid 21(3):31
axial velocity, and the enhancement of mixing efficiency is Damian IR, Hardt S, Balan C (2017) From flow focusing to vortex for-
not equally effective. mation in crossing microchannels. Microfluid Nanofluid 21(8):142
Different directions of future work may branch from the Dentz M, de Barros FP (2015) Mixing-scale dependent dispersion for
observations put forward in this article. One line of inves- transport in heterogeneous flows. J Fluid Mech 777:178–195
Galletti C, Arcolini G, Brunazzi E, Mauri R (2015) Mixing of binary
tigation is to use the toolbox of dynamical system theory fluids with composition-dependent viscosity in a t-shaped micro-
to single out a relevant length scale quantifying the aver- device. Chem Eng Sci 123:300–310
age thickness of the kinematic flux tube which is expected Gorodetskyi O, Speetjens M, Anderson P (2014a) Simulation and
to control the rate of homogenization onto the cross sec- eigenmode analysis of advective–diffusive transport in micromix-
ers by the diffusive mapping method. Chem Eng Sci 107:30–46
tion transversal to the average flow direction. This could be Gorodetskyi O, Speetjens MF, Anderson PD, Giona M (2014b) Analy-
achieved by extending to the three-dimensional setting the sis of the advection-diffusion mixing by the mapping method for-
methods developed in Meunier and Villermaux (2010) and malism in 3d open-flow devices. AIChE J 60(1):387–407

13
Microfluidics and Nanofluidics (2018) 22:20 Page 15 of 15  20

Gubanov O, Cortelezzi L (2010) Towards the design of an optimal Raynal F, Beuf A, Carrière P (2013) Numerical modeling of DNA-
mixer. J Fluid Mech 651:27–53 chip hybridization with chaotic advection. Biomicrofluidics
Hessel V, Löwe H, Schönfeld F (2005) Micromixers—a review on pas- 7(3):034107
sive and active mixing principles. Chem Eng Sci 60(8):2479–2501 Saad WS, Prudhomme RK (2016) Principles of nanoparticle formation
Lashgari I, Tammisola O, Citro V, Juniper MP, Brandt L (2014) The by flash nanoprecipitation. Nano Today 11(2):212–227
planar x-junction flow: stability analysis and control. J Fluid Mech Sadeghi A (2018) Micromixing by two-phase hydrodynamic focusing:
753:1–28 a 3d analytical modeling. Chem Eng Sci 176:180–191
Lasota A, Mackey MC (2013) Chaos, fractals, and noise: stochastic Singh MK, Kang TG, Meijer HE, Anderson PD (2008) The mapping
aspects of dynamics, vol 97. Springer, Berlin method as a toolbox to analyze, design, and optimize micromixers.
Leung AK, Tam YYC, Chen S, Hafez IM, Cullis PR (2015) Microflu- Microfluid Nanofluid 5(3):313–325
idic mixing: a general method for encapsulating macromolecules Sousa P, Coelho P, Oliveira M, Alves M (2011) Laminar flow in three-
in lipid nanoparticle systems. J Phys Chem B 119(28):8698–8706 dimensional square–square expansions. J Non-Newton Fluid
Liu H, Xu K, Zhu T, Ye W (2012) Multiple temperature kinetic model Mech 166(17):1033–1048
and its applications to micro-scale gas flows. Comput Fluids Stremler MA, Haselton F, Aref H (2004) Designing for chaos: applica-
67:115–122 tions of chaotic advection at the microscale. Philos Trans R Soc
Matsunaga T, Lee H-J, Nishino K (2013) An approach for accurate Lond A Math Phys Eng Sci 362(1818):1019–1036
simulation of liquid mixing in a t-shaped micromixer. Lab Chip Vikhansky A (2006) Coarse-grained simulation of chaotic mixing in
13(8):1515–1521 laminar flows. Phys Rev E 73(5):056707
Meunier P, Villermaux E (2010) The diffusive strip method for scalar Villermaux E (2012) On dissipation in stirred mixtures. Adv Appl
mixing in two dimensions. J Fluid Mech 662:134–172 Mech 45(91):2012
Oliveira M, Pinho F, Alves M (2012) Divergent streamlines and free Wong SH, Ward MC, Wharton CW (2004) Micro t-mixer as a rapid
vortices in Newtonian fluid flows in microfluidic flow-focusing mixing micromixer. Sens Actuators B Chem 100(3):359–379
devices. J Fluid Mech 711:171–191 Zhao C-X, He L, Qiao SZ, Middelberg AP (2011) Nanoparticle synthe-
Orsi G, Roudgar M, Brunazzi E, Galletti C, Mauri R (2013) Water-eth- sis in microreactors. Chem Eng Sci 66(7):1463–1479
anol mixing in t-shaped microdevices. Chem Eng Sci 95:174–183 Zhu T, Ye W (2010) Theoretical and numerical studies of noncon-
Raynal F, Beuf A, Plaza F, Scott J, Carrière P, Cabrera M, Cloarec tinuum gas-phase heat conduction in micro/nano devices. Numer
J-P, Souteyrand É (2007) Towards better DNA chip hybridization Heat Transf Part B Fundam 57(3):203–226
using chaotic advection. Phys Fluids 19(1):017112

13

You might also like