See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329858210

Viscous Effects on Inertial Drop Formation

Article  in  Physical Review Letters · December 2018

DOI: 10.1103/PhysRevLett.121.254501

CITATIONS READS
13 441

8 authors, including:

Antoine Deblais Ma Herrada

University of Amsterdam Universidad de Sevilla
35 PUBLICATIONS   99 CITATIONS    136 PUBLICATIONS   1,400 CITATIONS   

SEE PROFILE SEE PROFILE

Ines Hauner Krassimir Velikov

University of Amsterdam Unilever
7 PUBLICATIONS   38 CITATIONS    126 PUBLICATIONS   3,736 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cellular flows View project

Sloshing of magnetic liquids in microgravity (StELIUM) View project

All content following this page was uploaded by Antoine Deblais on 22 December 2018.

The user has requested enhancement of the downloaded file.

 PHYSICAL REVIEW LETTERS 121, 254501 (2018)

Viscous Effects on Inertial Drop Formation

A. Deblais,1 M. A. Herrada,2 I. Hauner,1 K. P. Velikov,1,3 T. van Roon,4 H. Kellay,5 J. Eggers,6 and D. Bonn1
1
Van der Waals-Zeeman Institute, Institute of Physics, University of Amsterdam, 1098XH Amsterdam, Netherlands
2
Depto. de Mecánica de Fluidos e Ingeniería Aeroespacial, Universidad de Sevilla, E-41092 Sevilla, Spain
3
Unilever R&D Vlaardingen, Olivier van Noortlaan 120, 3133 AT Vlaardingen, Netherlands
4
Technology Centre, University of Amsterdam, 1098XH Amsterdam, Netherlands
5
Laboratoire Ondes et Matiere d’Aquitaine, UMR 5798 CNRS-U. Bx, Universite de Bordeaux,
351 cours de la Liberation 33405, Talence, France
6
School of Mathematics, University of Bristol, University Walk, Bristol BS8 1 TW, United Kingdom
(Received 6 June 2018; revised manuscript received 10 September 2018; published 21 December 2018)

The breakup of low-viscosity droplets like water is a ubiquitous and rich phenomenon. Theory predicts
that in the inviscid limit one observes a finite-time singularity, giving rise to a universal power law, with a
prefactor that is universal for a given density and surface tension. This universality has been proposed as a
powerful tool to determine the dynamic surface tension at short time scales. We combine high-resolution
experiments and simulations to show that this universality is unobservable in practice: in contrast to
previous studies, we show that fluid and system parameters do play a role; notably a small amount of
viscosity is sufficient to alter the breakup dynamics significantly.

DOI: 10.1103/PhysRevLett.121.254501

Drop formation is a common phenomenon that occurs in trends have been confirmed by experiment and simulation
a broad range of industrial processes such as inkjet printing, [7,10,11], and Eq. (1) has been proposed as the basis of a
spraying, and cooling. In a typical laboratory experiment, a method to measure surface tension on a millisecond time-
drop is formed at the end of a syringe tip, and its detach- scale [12,13].
ment is filmed using a high-speed camera. Surface tension However, a careful analysis of past measurements in the
drives the flow that shrinks the diameter of the filament inertial regime reveals a considerable variation of the
connecting the drop to the syringe, and both fluid inertia prefactor A. Some of the values found in the literature
and fluid viscosity slows it down. The filament can be are 0.45–0.5 [10], 0.55 [14], 0.57–0.68 [7], or 0.97 [15].
observed to thin in a nonuniform way, which eventually
leads to the pinch-off of the drop (see Fig. 1) occurring at a
critical time tc . In the case of negligible viscosity, based on
the idea that the minimum neck radius Rmin is the only
characteristic length scale close to pinch-off, one predicts
the power law [1,2]

1=3
γ
Rmin ¼A ðtc − tÞ2=3 ; ð1Þ
ρ

where γ and ρ are the surface tension and the density of the
liquid, respectively.
Potential flow simulations and similarity theory [3–5]
confirm Eq. (1), with A ¼ 0.717… a universal constant
[4–6]. The profile near the pinch point converges toward a
double cone, one set inside the other with opening angles of
18° and 112° [4,7] [cf. Fig. 3(c) below]. This means that
eventually the profile must overturn [3,4,8], as indeed seen FIG. 1. High-speed image sequence (upper panel) of a drop of
water dripping from a faucet of 2 mm in radius. Lower panel:
in the last stages of pinch-off, shown in the last two panels
numerical simulation of the same problem using the Basilisk
of the simulations shown in Fig. 1. It is expected that the code, showing 2D cuts through the axis. Note the overturning of
above potential flow theory applies as long as the viscous the upper surface of the main drop in the last two frames. The
length scale lν ¼ ν2 ρ=γ is much smaller than Rmin [6,9], scale bar represents R0 ¼ 2 mm. The dimensionless time to
where ν is the kinematic viscosity of the liquid. These pinch-off τ ¼ ðtc − tÞ=t is indicated on the top of the figure.

0031-9007=18=121(25)=254501(5) 254501-1 © 2018 American Physical Society

 PHYSICAL REVIEW LETTERS 121, 254501 (2018)

Recently there has been a growing appreciation of the fact objectives Zeiss, 4× to 10× magnification and Navitar
that the approach to asymptotic power laws like Eq. (1) can 12×) recorded pinch-off events at up to 100 000 frames per
be slow [16–18], and may pass through one or several second with a spatial resolution as low as 1.6 μm=pixel. We
transient regimes before the final universal regime is paid particular attention to a stable setup, and used micro-
reached. scope objectives with very high contrast and low depth of
One obstacle to resolving the inconsistency in the field. Drop profiles were extracted using a homemade
measured values of A is that at the very last instant before MATLAB routine, from which Rmin ðtÞ was found.
the breakup, the drop profile overturns [7,19]. As seen in The numerical technique used to simulate the pinch-off
the cross sections produced from numerical simulation of low-viscosity pendant drops is a variation of that
(lower row of Fig. 1, 4th panel), the neck ends in a sharp tip, described in detail in Ref. [21]. It was previously applied
which lies inside an indentation formed on the top of the to study the effect of surface contamination on the
drop. The corresponding experimental image, taken from dynamics of pendant drops [22]. The main features of
the side, shows a projection onto the plane, which produces the method are the use of an analytical mapping to convert
a flat top of the drop, where overturning has occurred; the the numerical domain into a rectangular one and imple-
end of the neck is obscured. This means that the neck radius mentation of the Newton procedure to solve the nonlinear
can only be measured before overturning, at a typical size system of equations needed for a fully implicit treatment of
Rmin ≃ 5 μm [20]. the discretized equations. The numerical Jacobian required
Since the asymptotic value of A refers to the profile by the method is built with the help of analytical functions
having attained its final double cone shape after over- and the collocation matrix provided by the spatial discre-
turning occurs, conventional optical techniques are inca- tization of the mapped domain. With this method we can
pable of determining its asymptotic value. To circumvent accurately capture the dynamics of the pendant drops until
these optical limitations, and to probe length scales well the appearance of overturning.
below the optical resolution, Burton and Taborek [11] However, the analytical mapping fails after overturning
measured the electrical resistance r through the drop, and, like the experimental optical measurements, cannot be
whose main contribution comes from where the neck is used to study the subsequent dynamics. Therefore, closer to
thinnest, and thus scales like the resistivity ρr of the pinch-off and in order to compare with the electrical
material, divided by Rmin. The resistance was found to measurements in the mercury liquid bridge experiment,
scale in accordance with Eq. (1) down to nanometer length we used a numerical scheme based on a finite element
and nanosecond time scales. However, in order to calculate volume-of-fluid (VOF) method provided by the “Basilisk”
Rmin from r, the shape of the interface is needed. An software [23]. It uses QUADTREES [24] to allow efficient
approximate formula leads to a value of A ¼ 0.10 [11], adaptive grid refinement of the interface close to the pinch-
much smaller than the theoretical expectation or the result off region (see Supplemental Material, Sec. A [25]).
of previous optical measurements. The main results of our fluid pinch-off experiments are
In this Letter, to arrive at a definitive answer as to the presented in Fig. 2 for decane and water, using two different
validity of the scaling law (1), we use a combination of capillary sizes to assess the influence of the initial conditions.
experiment (using both optical and electrical measure- In the limit of slow dripping, the dimensionless parameters of
ments) and high resolution numerical simulations. We the experiment are the Bond number Bo¼ ρgR20 =γ and the
pffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
show that the correct asymptotic value of A is unobservable Ohnesorge number Oh¼ lν =R0 ¼ ν= ρR0 γ. We calcu-
for all practical purposes. On the one hand, the initial lated instantaneous values of A from the slope of
dynamics leaves its imprint and is slow to converge onto ðRmin =R0 Þ3=2 ¼ A3=2 τ as a function of the dimensionless
Eq. (1). On the other hand, the effect of viscosity is felt even qffiffiffiffiffiffiffiffiffiffiffiffiffi
if Rmin is greater than lν by several orders of magnitude, as breakup time τ ¼ ðtc − tÞ=t , with t ¼ ρR30 =γ the capil-
is usually the case, e.g., for water, with lν ¼ 14 nm. lary time. This procedure does not involve a determination of
Different liquids allow us to perform pinch-off experi- the extrapolated time to pinch-off tc for the determination of
ments with different ratios γ=ρ. We used decane from the prefactor A, and therefore does not make an assumption
Sigma Aldrich and highly ultrapure water from a Millipore on the validity of Eq. (1).
Milli-Q system. Surface tensions γ were verified independ- Surprisingly, for both liquids, over the range 2 × 10−3 <
ently using the pendant drop method (Kruss Easy Drop); τ < 1 before overturning (indicated by a cross), the
we used γ water ¼ 72 mN m−1 and γ decane ¼ 24 mN m−1 . A prefactor varies nonmonotonically between 0.4 and 0.6.
syringe pump (Harvard Apparatus) supplied the solutions In addition, our mapping simulations perfectly capture the
at the needle tip at a slow rate (needle tip radii: R0 ¼ 0.2 experimental variation of the prefactor in all the cases
and 2 mm), so we can observe the detachment of a single presented here until overturning. To follow the dynamics of
drop at a time. thinning beyond the point of overturning, we used Basilisk
A high-speed camera (Phantom V701, Vision Research) simulations, which agree very well with data before over-
equipped with a microscope tube lens (Microscope turning. Although Rmin is still greater than lν by more than

254501-2
 PHYSICAL REVIEW LETTERS 121, 254501 (2018)
(a)

(b)

FIG. 2. (a) Prefactor A as a function of the dimensionless time to pinch-off τ for water and decane for different values of the capillary
size R0 , the Ohnesorge number Oh, and Bond number Bo. Symbols indicate experimental results whereas solid lines were obtained from
two different simulation techniques: mapping (red) and Basilisk (black). After a nonmonotonic variation, A reaches a maximum value
and starts to decrease on account of viscous effects. The theoretical prediction for A is depicted as a horizontal dashed line. (b) The
prefactor A is plotted as a function of the characteristic viscous dimensionless timescale ðtc − tÞ=tν : all the curves (from basilisk
simulation) collapse into a single master curve for late times, showing that the drop of the prefactor results from viscous effects.

2 orders of magnitude, A starts to decrease on account of entire setup is placed in a chamber flushed continuously
viscous effects (as an example, see the case of Decane for with pure nitrogen. In fact, we believe surface contamina-
large capillary at τ ∼ 10−3 in Fig. 2). To confirm that the tion to be of minor importance for the asymptotic regime,
drop of the prefactor as a result of viscous effects is as the new surface is created faster than impurities can
physical, we plot in Fig. 2(b) the prefactor A as a function arrive [28]. The maximum bandwidth of our oscilloscope
of the dimensionless time ðtc − tÞ=tν , where tν ¼ ν3 ρ2 =γ 2 (RTO 1024, Rohde and Schwartz) is limited to 2 GHz, with
is the characteristic viscous timescale. We find that this a sampling rate of 10 × 109 samples/s. The high-frequency
collapses the drop of the prefactor A for late times, response of our system is evaluated by bringing the two
confirming the physical origin of this behavior. Thus even copper electrodes together in the absence of mercury,
for the smallest Ohnesorge numbers the prefactor never which gives a step function with a rise time of 300 ps.
reaches the asymptotic value of 0.70 reported in Ref. [10]. The recorded oscilloscope data represent the output voltage
This observation is well confirmed in the Supplemental V s ðτÞ of the voltage divider circuit [11], which decreases
Material [25], where we simulate two artificial liquids with as the fluid neck becomes thinner [cf. Fig. 3(b)]. The
the same control parameters except for Oh, which is varied resistance rHg of the mercury bridge is found from
by almost 1 order of magnitude (see Supplemental Material rHg ¼ rscope ðV in =V s − 1Þ, with rscope ¼ 50 Ω the internal
[25], Sup. Fig. 3). resistance of the scope and V in ¼ 1 V the tension supplied
To see whether the asymptotic state can be seen by the voltage generator (see Supplemental Material [25]).
experimentally, we push our experimental investigation To compare the measured r to numerical simulation, we
closer to pinch-off by measuring the resistance of a mercury numerically calculate the current flow through the com-
bridge [see Figs. 3(a),3(b)]. We use mercury for its low puted profile, which is shown in Fig. 3(c) for different
resistivity ρr and on account of its small viscous length values of Rmin and compared to the profile of the inviscid
scale lν ¼ 4.2 × 10−10 m, even smaller than that of water. similarity solution [5]. At Rmin ¼ 10−4 the computed
Our electrical circuit consists of a voltage-biased voltage profile comes very close to the similarity profile, but then
divider [11]. A mercury capillary bridge is formed between overshoots to form an even larger angle than predicted on
two copper electrodes (radius R0 ¼ 1 mm); the upper one the drop side. To compute the resulting electrical resistance,
can be translated vertically with a step motor. A droplet of we solve Laplace’s equation inside the liquid, with
volume V 0 ¼ 5 μL of fresh mercury is deposited on the Neumann boundary conditions on the surface [29] (see
lower electrode. Prior to measurements, the mercury sur- Supplemental Material, Sec. B [25] for more details). The
face is cleaned with concentrated sulfuric acid (H2 SO4 ). resulting dimensionless resistance is shown in Fig. 3(d),
To prevent any further oxidation of the superficial layer and indeed continues to evolve down to the smallest scales
of mercury with air during the capillary bridge breakup, the considered. Owing to the continuous evolution of the

254501-3
 PHYSICAL REVIEW LETTERS 121, 254501 (2018)
(a) (c) (e)

(b) (d) (f)

FIG. 3. Breakup dynamics of a mercury bridge (Oh¼ 6.0437 × 10−4 ) from the millimeter to the nanometer scale. (a) Electrical
resistance of a mercury capillary bridge between two electrodes as function of time. The upper electrode moves upwards, initiating
breakup; the bridge’s spatial profile is recorded simultaneously using high-speed imaging. (b) Typical measurement of voltage V s over
the divider circuit as a function of time during the breakup process (note the nanometer timescale); at breakup (denoted here as t ¼ 0) the
voltage decreases suddenly. This is used to trigger fast imaging. (c) The shape of the liquid bridge RðzÞ is computed from a Basilisk
simulation, allowing us to calculate the resistance r of the filament in time. (d) The resistance as function of minimum radius, compared
to the asymptotic value of the inviscid solution, and the estimate by Burton and Taborek. (e) The measured resistance of the liquid bridge
data is well described by the simulations very close to the pinch-off point. (f) The dynamics of thinning of the liquid bridge before
overturning recorded by imaging techniques is also well captured by Basilisk simulations. In inset, the prefactor A of mercury is plotted
as a function of the dimensionless time to pinch-off τ.

self-similar profile, the resistance does not converge to the provides the necessary evidence that the numerics work well.
value r ¼ 1.51ρr =Rmin found for the inviscid similarity Thus, taken together, the measurements and simulations give
solution, and is considerably higher than r ¼ 0.97ρr =Rmin the results for the prefactor A over a large range of length
estimated in Ref. [11]. We therefore compare directly the scales and timescales [see inset of Fig. 3(f)]. The comparison
simulated resistivity with that obtained by the experiment. also shows that for low viscosity fluids such as mercury
The electrical setup is coupled to the fast camera, (Oh ¼ 6.0437 × 10−4 ), the prefactor remains close to the
triggered by the oscilloscope of the breakup event; this purely inviscid prediction until the effects of viscosity come
allows us to compare electrical with optical measurements into play.
in the same experiment. The optical measurement works In conclusion, we study drop pinch-off in cases where
over the range 0.05 ms < t < 0.1 s and 10 μm < the inner (viscous) length scale is 5 (water) or 7 (mercury)
Rmin < 1 mm, while the electrical measurements have orders of magnitude greater than the outer length scale R0 ,
much higher resolution: 1 ns < t < 0.7 μs and 10 nm < both experimentally and computationally. We find that the
Rmin < 10 μm. Our electrical resistance measurements are dimensionless prefactor A exhibits a complex, nonmono-
presented in Fig. 3(e) and show quantitative agreement with tonic behavior over many orders of magnitude in time,
the simulation on timescales well below any where such a measured relative to the breakup time. We never found the
comparison has been made so far. Only farther away from prefactor to fully reach the asymptotic value of A ¼ 0.717.
pinch-off does our resistance measurement deviate from This indicates the existence of very slow transients, similar
simulation, as other electrical resistance effects start come to transients reported for more viscous fluids [17,18].
into play when the resistance is no longer dominated by the However, in those experiments the theoretical value of
neck region alone. However, for longer times we can use the prefactor is reached eventually both for the viscous and
the optical measurement done in the same experiment; these viscous-inertial regimes [18], indicating that in our experi-
also agree perfectly with the simulations [cf. Fig. 3(f)]. ments the inertial transients are even slower and the
The electrical experiments alone do not allow us to unam- asymptotic value is not even reached for nanometric length
biguously show the nonuniversality since the neck shape and nanosecond time scales; for smaller length scales and
is not known a priori. The simulations do provide the neck timescales one may even wonder whether a continuum
shape and by comparing to the experiment these allow us to hydrodynamics description remains valid. Apart from the
unambiguously determine the radius at each time and there- fundamental interest, the full understanding of the slow
fore the prefactor A. The agreement with the experiments time dependence of A is crucial to measure the dynamic

254501-4
 PHYSICAL REVIEW LETTERS 121, 254501 (2018)

surface tension of liquids: if one wishes to determine this, [13] I. M. Hauner, A. Deblais, J. K. Beattie, H. Kellay, and D.
one has to take into account that fluid and system Bonn, J. Phys. Chem. Lett. 8, 1599 (2017).
parameters play an important role. [14] M. P. Brenner, J. Eggers, K. Joseph, S. R. Nagel, and X. D.
Shi, Phys. Fluids 9, 1573 (1997).
We are grateful to David Leppinen for sharing numerical [15] J. C. Burton, J. E. Rutledge, and P. Taborek, Phys. Rev. E
data of similarity profiles from Ref. [5]. J. E. acknowledges 75, 036311 (2007).
support from the Leverhulme Trust through International [16] J. Eggers, M. A. Fontelos, D. Leppinen, and J. H. Snoeijer,
Academic Fellowship IAF-2017-010. H. K. thanks the Phys. Rev. Lett. 98, 094502 (2007).
Institut Universitaire de France for partial support. [17] J. R. Castrejón-Pita, A. A. Castrejón-Pita, S. S. Thete, K.
Sambath, I. M. Hutchings, J. Hinch, J. R. Lister, and
M. A. H. thanks the Ministerio de Economía y competiti-
O. A. Basaran, Proc. Natl. Acad. Sci. U.S.A. 112, 4582
vidad for partial support under the Project No. DPI2016- (2015).
78887-C3-1-R. This project has been partially supported in [18] Y. Li and J. E. Sprittles, J. Fluid Mech. 797, 29 (2016).
part by Unilever R&D Vlaardingen. [19] J. C. Burton, J. E. Rutledge, and P. Taborek, Phys. Rev. Lett.
A. D. and M. A. H. contributed equally to this work. 92, 244505 (2004).
[20] J. C. Burton and P. Taborek, Phys. Fluids 19, 102109
(2007).
[21] M. A. Herrada and J. M. Montanero, J. Comput. Phys. 306,
[1] D. H. Peregrine, G. Shoker, and A. Symon, J. Fluid Mech. 137 (2016).
212, 25 (1990). [22] A. Ponce-Torres, J. M. Montanero, M. A. Herrada,
[2] J. Eggers, Phys. Rev. Lett. 71, 3458 (1993). E. J. Vega, and J. M. Vega, Phys. Rev. Lett. 118, 024501
[3] Y.-J. Chen and P. H. Steen, J. Fluid Mech. 341, 245 (1997). (2017).
[4] R. F. Day, E. J. Hinch, and J. R. Lister, Phys. Rev. Lett. 80, [23] http://basilisk.fr/.
704 (1998). [24] S. Popinet, Ocean Dyn. 61, 1261 (2011).
[5] D. Leppinen and J. R. Lister, Phys. Fluids 15, 568 (2003). [25] See Supplemental Material at http://link.aps.org/
[6] J. Eggers and E. Villermaux, Rep. Prog. Phys. 71, 036601 supplemental/10.1103/PhysRevLett.121.254501 for details
(2008). on the validation of the numerical codes, numerical comput
[7] J. R. Castrejon-Pita, A. A. Castrejon-Pita, E. J. Hinch, J. R. ance in the liquid bridge, and more details on the electrical
Lister, and I. M. Hutchings,Phys. Rev. E 86, 015301(R) setup, which includes Refs. [19,26,27].
(2012). [26] C. Ferrera, M. A. Herrada, J. M. Montanero, M.
[8] P. Lenard, Ann. Phys. (Berlin) 266, 209 (1887). Torregrosa, and V. Shevtsova, Phys. Fluids 26, 012108
[9] J. R. Lister and H. A. Stone, Phys. Fluids 10, 2758 (1998). (2014).
[10] A. U. Chen, P. K. Notz, and O. A. Basaran, Phys. Rev. Lett. [27] M. A. Herrada, J. M. López-Herrera, E. J. Vega, and J. M.
88, 174501 (2002). Montanero, Phys. Fluids 23, 012101 (2011).
[11] J. C. Burton, J. E. Rutledge, and P. Taborek, Phys. Rev. Lett. [28] M. Roché, M. Aytouna, D. Bonn, and H. Kellay, Phys. Rev.
92, 244505 (2004). Lett. 103, 264501 (2009).
[12] M. R. De Saint Vincent, J. Petit, M. Aytouna, J. P. Delville, [29] W. R. Smythe, in Static and Dynamic Electricity (McGraw-
D. Bonn, and H. Kellay,J. Fluid Mech. 692, 499 (2012). Hill, New York, 1950).

254501-5

View publication stats