Effect of inertia on the dynamic contact angle in oscillating menisci

Domenico Fiorini,1, 2, a) Miguel Alfonso Mendez,1 Alessia Simonini,1 Johan Steelant,3, 4 and David Seveno2
1) vonKarman Institute for Fluid Dynamics, Waterloosesteenweg 72, Sint-Genesius-Rode,
Belgium
2) KU Leuven, Dept. of Materials Engineering, Leuven 3001, Belgium
3) ESTEC-ESA, Keplerlaan 1, Noordwijk, The Netherlands
4) KU Leuven, Dept. of Mechanical Engineering, Leuven 3001, Belgium

(Dated: 11 August 2022)

The contact angle between a gas-liquid interface and a solid surface is a function of the dynamic conditions of the
contact line. Classic steady correlations link the contact angle to the contact line velocity. However, it is not clear
whether they hold in presence of inertia and in the case of perfect wetting fluids. We analyze the shape of a liquid
arXiv:2208.05329v1 [physics.flu-dyn] 10 Aug 2022

interface and the corresponding contact angle in accelerating conditions for two different fluids, i.e. HFE7200 (perfect
wetting) and demineralized water. The set-up consists of a U-shaped quasi-capillary tube in which the liquid column
oscillates in response to a pressure step on one of the two sides. We obtained the evolution of the interface shape from
high-speed back-light visualization, and we fit interface models to the experimental data to estimate the contributions
of all the governing forces and the contact angle.
Traditional interface models fail to predict the interface shape and its contact angle at large interface and contact line
acceleration. We propose a new model to account for the acceleration, and we discuss its impact on the measurement
of the transient contact angle.

I. INTRODUCTION ize poorly. Petrov and Sedev 25 consider the dynamic contact
angle as "inaccessible" by optical means and attempt to over-
The curved shape of a gas-liquid interface produces a pres- come this limit by computing the contact angle from the the-
sure difference across the interface itself. This is known as oretical quasi-static interface shape fitted to the experimental
capillary pressure and plays a key role in the modeling of the data. This theoretical interface shape is obtained by solving
interface motion when the gravity and the capillary forces be- a boundary value problem which Petrov and Sedev 25 closed
come comparable1,2 . This is the case of many practical ap- using Neumann boundary conditions, i.e. the interface slope
plications such as the imbibition of fluids by porous media3 , on the wall and far from the solid surface.
enhanced oil recovery operations4 , droplet manipulation in Methods for measuring the contact angle using an analytic
microfluidics2,5,6 and liquid management in space systems7 . formulation of the meniscus interface are referred to as Menis-
In all of these problems the dynamic contact angle, i.e., the cus Profile Methods (MPM)26 . Maleki et al. 27 extended the
angle formed by the interface at a moving contact line, rep- approach of Petrov and Sedev 25 to a "quasi-steady" approx-
resents the boundary condition for the interface shape and imation where the shape of the gas-liquid interface depends
consequently governs the role of the capillary pressure of the on the capillary number. Iliev and Pesheva 26 uses the same
system8 . approximation, closing the boundary value problem using the
The first definitions of dynamic contact angle was given by interface locations near the wall, available experimentally. In
Blake, Hoffman and Voinov9–11 . These studies show that the the case of receding interfaces, Maleki et al. 27 found a crit-
motion of the contact point and the contact angle are strongly ical capillary number above which the contact angle seem-
correlated. In the case of viscous dominated flows9–19 , the ingly jumps to zero and a liquid film is deposed along the
dynamic contact angle can be predicted from the Capillary walls. Voinov and Eggers28,29 also hypothesized the existence
number Ca = µuc /σ , where µ is the dynamic viscosity of the of such condition.
liquid, uc is the contact line velocity and σ is the surface ten- The case of inertia-dominated interfaces has not been con-
sion between gas and liquid. Both theoretical and empirical sidered in the framework of MPM methods. Inertia can have
correlations were developed for the case of steady contact line an important role in classic problems such as the capillary
motion. However, several authors20–24 have shown that these rise, wherein inertia and surface tension control the initial
approaches fail in inertia dominated conditions, and more ex- accelerating phase. The typical modelling approach for this
perimental data is required to clarify the role of contact line problem (see Quéré 22 ) considers a quasi-steady formulation
acceleration and the history of the contact line dynamics20,21 . for the contact angle and assumes that the interface remains
Experimental studies on dynamic contact angles emphasize spherical also in dynamic conditions. These models have been
the difficulty to obtain reliable results due to the microscopic proved valid for high viscous fluids over a large range of Ca
scales involved and thus the high resolution needed for the (see22,30 ) and in microgravity conditions31 . These have also
optical equipment to visualize the interface. Consequently, been used to describe the results from experiments in short
the results are often affected by high uncertainty and general- tubes23 , i.e. for which liquid can spread onto the external
walls after reaching the top of the tube, and the inertial uptake
of liquid drops24 , for which the capillary suction due to the
meniscus interacts with the Laplace pressure in the (shrink-
a) Electronic mail: [email protected] ing) drop.
 2

However, as also reported by Quéré 22 , these models are rinsed with an isopropanol solution and let dry before closing
incapable of describing capillary rise experiments with less all the connections with the laboratory environment. After the
viscous fluids, especially in the presence of oscillations of the tests with HFE7200, the tube was emptied, dried at ambient
liquid column. Whether the discrepancy is due to the incorrect conditions, and prepared for the test with water using the same
modelling of the interface shape, the miss-prediction of the procedure.
contact angle evolution, the modelling of viscous forces or the Initially, the U-tube is filled with the test liquid while the
modeling of the pressure drop at the tube entrance remains an fast response valve connecting the two sides is open, and the
open question. two interfaces of the liquid column have the same height. The
In this work, we analyze two of the aforementioned mod- initial conditions of the experiment are defined by closing the
elling challenges, namely the description of the gas-liquid in- fast response valve and increasing the pressure on side A. This
terface and the contact angle dynamics. The chosen config- sets the initial level difference between the two interfaces and
uration is a U-tube geometry, where an axial-symmetric gas- zero velocity of the interface.
liquid interface forms on the two sides. The tube radius was
selected to maximize the impact of capillary forces on the in-
terface force balance. We measure the dynamic contact an-
gle using an adapted MPM and introduce an empirical term
to account for the impact of inertia in circular tubes. We ana-
lyze the interface dynamics for two low viscous fluids, namely
demineralized water and HFE 7200 by 3M Novec.
The experimental configuration is described in section II.
The interface parametrization is described in section III. Re-
sults are presented in section IV while section V closes with
conclusions and outlooks to future works.

II. EXPERIMENTAL METHODS

Table I shows the relevant properties for the fluids con-

sidered in this study, namely demineralized water and pure
HFE7200 (C4 F9 OC2 H5 ) produced by 3M Novec. The
HFE7200 is a synthetic liquid in the family of hydrofluo-
roethers characterized by high volatility (vapor pressure 14
kPa at 25◦C) and small equilibrium contact angles (below 10
degrees). These liquids are frequently used to simulate the
properties of cryogenic space fluids32 .
FIG. 1: Left: schematic diagram of experimental setup. At
TABLE I: Fluids physical properties at 298 K initial conditions, side A is pressurized and side B is at room
pressure, a fast response valve allows for suddenly releasing
Water HFE7200 the pressure difference to begin the dynamic experiment.
Right: example of complete image acquired.
density (ρ) kg/m3 996.66 1423.00
dynamic viscosity (µ) mPa · s 0.97 0.64
The preparatory phase ends by closing all connections
surface tension (σ ) mN/m 72.01 13.62
to the external environment and waiting sufficient time for
HFE7200 to saturate the gas volumes within the tube. The
The behavior of the gas-liquid interface of HFE7200 and experiment starts by opening the fast-response valve, and
water is analyzed in a U-shaped quartz tube manufactured by the two liquid interfaces start oscillating around the equilib-
PIERREGLAS (Belgium). The U-tube internal diameter is rium position. The motion of one of the two interfaces is
8 mm with axial length 164 mm (see Figure 1). The two recorded with a high-speed camera SP-12000-CXP4 (STEM-
sides of the tube (A and B in Figure 1) are connected by MER IMAGING), acquiring grey-scale images at 300 fps.
flexible pipes. A fast response valve controls the connection The interface shape is visualized by a diffused light source
between these sections. Two differential piezo-resistive mea- on the opposite side of the tube. The active region of the cam-
suring cells (AMS 5812, ANALOG MICROELECTRONICS) era sensor is restricted to the central portion with 4096x512
are placed at the two entrances of the tube. The Side A is con- pixels to achieve the highest acquisition frequency allowed
nected to an input pressure line, while the side B is connected by the camera. The camera mounts a Nikon micro-Nikkor
to a discharge pressure line. The first experimental campaign 105mm lens and is positioned to visualize the motion of the
was carried out with HFE7200. interface while spanning the full tube width, as shown in Fig-
The tube was kept as provided protected from the external ure 1. Each frame is analyzed with image processing tech-
environment until the test. Then, the tube internal surface was niques and the interface shape is converted in a set of points
 3

(ri , hi ) with i ∈ [1, n p ] and n p the number of points detected.

Details about the image processing and the correction for op-
tical distortions are given in the supporting information.



>PP@
A. Image analysis

The image processing for the interface detection consists
of four main steps, recalled in Figure 2. In the first step, 
the original images (Figure 2a) are filtered via global noise
removal based on n-local means denoising algorithm33 , fol-      
lowed by a first image recontrasting (Figure 2b). The second >PP@
and third steps process each column of the image separately.
The grayscale profiles are first filtered using a Savitzky–Golay FIG. 3: Example image of HFE7200 interface without
filter34 and the resulting image re-contrasted using directional correction (red circles) and with correction (white circles).
grayscale averaging35 . This allows for increasing the grey The vertical lines corresponds to the real inner surface of the
level intensity in the region of the liquid interface close to the tube.
tube walls, which appear darker compared to the centre of the
channel. 
5HJUHVVLRQ
   'DWD

   
>PP@

>PP@

>PP@

   h/lc>@ 

            
>PP@ >PP@ >PP@
(a) Original. (b) Recontrasted. (c) Gradients.

 
$  
  
   U5>@
>PP@

>PP@

>PP@

   FIG. 4: Corrected HFE7200 interface profiles and regression
with static interface with lc = 0.99mm and θS = 8.6.
           
>PP@ >PP@ >PP@
(d) Edges. (e) Interface. (f) Result.

FIG. 2: Sequence of image processing steps implemented to

identify the liquid interface
FIG. 5: Side view of HFE7200 droplet (left) and water
droplet (right) for static contact angle quantification. x =0
coincides with the apex of the droplet while y = 0
B. Optical correction corresponds to the positioning of the contact point within the
image. The y-axis has a different scale on the left and right
The circular shape of the tube deforms the image of the side.
meniscus in the radial direction. This implies that the scaling
of the image follows a non linear trend and the size of one
pixel in the radial direction varies from 21 µm in the channel tortion is maximum at the contact point of the meniscus and
center to 9 µm in the channel wall. The size of one pixel is disappears at the centre of the channel. Following Darzi and
21 µm along the channel axis. The meniscus appears larger Park 36 , the distortion can be corrected given the distance of
than its actual size, and the contact point with the solid surface the camera from the tube center and the refraction index of
seemingly lies in the region between the inner and outer side the materials involved. We validate the procedure by compar-
of the tube, as shown by the red points in Figure 3. The dis- ing the corrected profiles with the theoretical profile in static
 4

conditions (see Eq. 3 in section III A). A similar approach is

also used by Petrov and Sedev 25 . In static conditions, the in-
terface shape is determinedp by the static contact angle θS and
the capillary length lc = σ /ρg.
The regression of the static profiles with the static menis-
cus model (see Figure 4) yields lc = 0.98 ± 0.01 mm and
θs = 6.57◦ ± 1.48◦ for HFE7200, lc = 2.72 ± 0.1 mm and
θs = 22.4◦ ± 1.6◦ for demineralized water. The capillary
length computed from the liquid properties at the laboratory
temperature (25◦ , see Table I) is lc = 0.98 mm for HFE7200
and lc = 2.71 mm for demineralized water. Moreover, the
static contact angle obtained from the side view of HFE7200
and water droplets (see Figure 5) deposited on a flat sub-
strate made of an equally prepared quartz material yields
θs = 7.1◦ ±2◦ for HFE7200 and θs = 20.33◦ ±1◦ for deminer-
alized water. This validates both the static model formulation
and the optical correction implemented in this work.
As expected, the corrected profile (white circles in Figure FIG. 6: Schematics of the interface model. The origin is
3) lies inside the tube and wets its inner surface (vertical white located at the interface, in the center of the channel, and
lines in Figure 3). moves with it, i.e. h(0,t) = 0 for all times.

III. CONTACT ANGLE MEASUREMENT having used the short hand notation hr and hrr to denote the
first and second partial derivatives of h with respect to the ra-
Both water and HFE7200 wet the inner walls of the quartz dial coordinate r. Eq. 1 was solved with Neumann conditions
tube. Several authors have shown that the simple Tangent- at the wall and Dirichlet conditions at the channel centre. Thus
Line methods (TLM), i.e. a linear extrapolation of the gas- the problem reads:
liquid interface to the wall, tends to overestimate the con- 
tact angle25–27 , especially with perfect wetting liquids like ∇ · n + lc−2 h = 0
HFE7200. h (R,t) = ctg(θ (t)) , (3)
 r
The MPM used in this work retrieves the contact angle by hr (0,t) = 0
fitting a model for the interface to the available data. This
having introduced the capillary length lc , and denoting as θ (t)
model depends on various parameters, among which the con-
the dynamic contact angle, defined as the "apparent dynamic
tact angle. The contact angle measure is thus cast as a re-
contact angle" by Petrov and Sedev 25 . Model (3) provides
gression problem and the accuracy of the method depends on
a good approximation of the interface shape also for small
the accuracy of the interface model. We review three inter-
contact line velocities, and it is often called quasi-static25 . In
face models in this section. The first two are the static and
these conditions, the viscous bending of the interface near the
quasi-steady models (III A), used by various authors25–27,37 .
contact line is not visible at a ‘macroscopic’ scale. i.e. at the
We then move to a new approach to account for the inertia
scale of the "apparent" contact angle37 .
contribution in subsection III B.
A simple analytic solution can be derived under the as-
sumption that hr  1 and thus ∇ · n ≈ hrr . In what follows, we
do not rely on this simplification and solve (3) numerically. It
A. Inertial-less modeling
is worth noticing that this model has a singularity at r = 0,
as the curvature approaches zero. This was circumvented by
Figure 6 illustrates the definition of the interface model and excluding the origin in the computational domain.
the relevant parameters. At rest, the shape is controlled by the When the capillary number is not negligible, the viscous
balance of gravitational and capillary forces38 . The resulting forces inside the meniscus must be included in the force
interface is axial symmetric, and the force balance in an in- balance26,27 . At a constant contact line velocity uc , the viscous
finitesimally thin control volume bounded by the interface at pressure can be modeled with the Voinov approximation11 :
y = h(r) and y = 0 reads:
µUcl
∆pν = 3 F(β ) , (4)
σ ∇ · n + ρgh = 0 (1) (R − r)

where n is the unit normal vector and ∇ · n is the interface where F(β ) is a correction factor39 for small slopes of the
curvature. For an axial-symmetric meniscus, these are: interface, and β = cot(hr ) is the interface slope. This term
can be expressed as F(β ) = sin(β (r))/(β (r)2 ). In case of
z − hr r −rhr − r2 hrr receding contact lines, this approximation holds until a critical
n= ∇·n = , (2)
1 + h2r r2 (1 + h2r )3/2 capillary number28,40 , above which the contact line vanishes
 5

and it is impossible to define a contact angle. The ‘quasi- set of parameters w(t). We have w(t) := θ (t) for the grav-
steady’ model is thus a correction of (3) to include (4): ity based model (3) and the viscous gravity model (5) and
w(t) := θ (t), ct (t), li (t) for the inertia-corrected model (7).
∇ · n + lc−2 h − 3 (R−r)
Ca

F(β ) = 0 The regression aims at finding the set of parameters w(t)
hr (R,t) = ctg(θ (R,t)) . (5) which allow each model h(r; w(t)) to best approximate the

hr (0,t) = 0 experimental points (ri , hi ) sampling the interface shape, with
i ∈ [1, n p ] and n p the number of points. This allows for indi-
This model is valid for quasi-steady (i.e. with moderate rectly obtaining the contact angle at each time step.
accelerations) motion of the contact line, where the receding To evaluate the accuracy of the interface regression, we use
capillary number is below the critical value, and the flow field the average of the squared Fréchèt distances between the set
in the meniscus is dominated by pressure and viscous forces of points and the regressed curves. For each of the points
(lubrication approximation). (ri , hi ), the Fréchèt distance to the interface h(r; w) is denoted
Because (5) has also a singular point at the wall, the numer- as (dε (xi , yi , h(ri , w)). This is the radius of the smallest cir-
ical domain does not reach r = R, but stops at r = R − ld . The cle, centered in (ri , hi ) and tangent to the curve h(r; w). The
exact value of ld is under debate, however most authors11,27,39 use of the Fréchèt distance allows for giving equal importance
place it in the range 10−5 − 10−6 mm. to errors and uncertainties along both axes, and this is essen-
tial in regions near the wall, where θ → 0 and thus |hr | → ∞.
More generally, the Fréchèt distance is particularly useful for
B. Correction for inertia effects regressing functions where the modulus of the derivative is
large (see e.g.41 ).
Although the lubrication approximation is valid in the The best set of weights, for each model, is thus the one that
vicinity of the contact line for a large range of conditions, at minimizes the objective function
some distances from walls, the flow inertia impacts the menis-
cus shape. Given the difficulties in modelling this effect from n
1 p 2
first principles, we account for its impact with an additional argmin J(w) = ∑ dε (ri , hi , h(ri ; w) . (8)
(heuristic) pressure term of the form: w n p i=1

Figure 7 shows an example of interface regression with the

 
− r−R
Ha (r,t) = ρai (t)lh 1 − e i l , (6) circles in each point illustrating the local Fréchèt distance.
The inset allows for better appreciating the accuracy of the
where ai (t) is the acceleration of the interface as a function of fit near the wall.
time, lh = Rct is a characteristic length defined by the model
parameter ct , and li (t) controls how the contribution of inertia
decays towards the tube wall. This terms plays the role of a  0RGHO
dynamic pressure, arising as the accelerating liquid column 'DWD 

adapts to the interface velocity within a certain distance from
it. Its mathematical form has been chosen to allow a wide  
variety of interface shapes depending on the closing parameter
KOF>@

li (t), to be identified via regression of the interface shape.    
For small interface accelerations, Ha is negligible and the
quasi-steady form of the equation is retrieved. Summarizing, 
the inertia-corrected model has the form:

∇ · n + lc−2 h − 3 (R−r)
Ca

F(β ) + Ha (r,t) = 0
. (7) 
hr (R,t) = ctg(θ (R,t))
hr (0,t) = 0      

U5>@
In section IV, we refer to system (3) as gravity-based (GB)
model, to system (5) as viscous-gravity (VG) model, and to FIG. 7: Visualization of the regression based on Fréchèt
system 7 as Inertia-corrected (IC) model. distance in experimental image. The radius of each circle
corresponds to the Fréchèt distance dε, j of the data point
(filled circle) from the model (solid line). The regression
C. Interface regression aims at finding the model parameters that minimize the
function J(w) in (8).
An optimization routine is used to solve the regression of
the interface shape with the different interface models. Let The optimization was carried out using the Nelder-Mead
h(r; w(t)) (see Figure 6) denotes the interface profiles solv- algorithm42 , implemented in Python library scipy.optimize43 .
ing problems (3), (5) or (7) at time t, parametrized by the For each model, we used the boundary value problem solver
 6

in scipy.integrate43 ; this implements a collocation algorithm 

coupled with a damped Newton method44 . The optimization
space for the weights was identified by trial and error and the
optimization is carried out on a normalized domain, i.e. keep- 
ing all weights in the range [0, 1] using a min-max scaling.
Uncertainties are computed using a bootstrapping
approach45 . For each experimental image, the regression 

out>GHJ@
is repeated nb = 500 times using 70% of the available n p
points. Each time, the points used in the regression are
sampled randomly. The results of the multiple regressions 
produce a distribution of weights w ∼ p(w). Since these were
approximately Gaussian, the uncertainty in each parameter

is computed as 1.96 times the (sample) standard deviation of
the weight distribution, thus considering a 95% confidence 030hm = 1 m, 10 m, 30 m
7/0hm = 1 m, 10 m, 30 m
interval. The uncertainty in the interface detection is then 
computed using a Monte Carlo approach: the distributions      
in>GHJ@
of weights are propagated by solving the model equations nb
times and computing the associated uncertainty bounds from FIG. 8: Accuracy comparison between two different
the standard deviations of the corresponding interface h(r; w) methods, MPM (circles) and TLM (squares), for contact
distributions. angle quantification on synthetic images by using different
degree of vicinity to the contact line: lm = 1µm (full
symbols), lm = 10µm (left-filled symbols) and lm = 30µm
D. Validation of the image analysis method (empty symbols).

We analyzed the accuracy of the proposed MPM method by

testing it on a set of meniscus profiles for which the ground- reports, with red squares, the results obtained using the TLM.
truth is available. To this end, we use the IC model (7) to The result shows that the TLM significantly over-predicts
generate a set of profiles by varying the model parameters the contact angle when this is below 15◦ ; this result is in line
w := ct , li , θ and Ca within the ranges observed in the ex- with those reported by Maleki et al. 27 . The interface reso-
periments. This means ct ∈ [−1, 1], li ∈ [4.e − 4, 4.e − 3] m, lution’s impact becomes more pronounced at θ < 15◦ , while
θ ∈ [5◦ , 45◦ ] and Ca ∈ [−5.e − 3, 5.e − 3]. We randomly sam- the proposed MPM is more accurate and more robust against
pled 30 points using a uniform distribution within these ranges poor interface resolution over the entire range of parameters
and generate the associated interface profiles h(r; w,Ca) by and analyzed operating conditions. Nevertheless, it is worth
solving (7) on a domain r ∈ [0, (R−lm )], where lm is the small- noticing that the uncertainties in the contact angle determina-
est distance from the wall at which it is possible to sample the tion grow with smaller contact angles and reach up to 5◦ in
interface location. Noticing that in our experiments lm varies some conditions, regardless of the interface resolution.
in the range [1, 5] µm, we consider lm = (1, 10, 30) µm, to ana-
lyze the impact of the image resolution in the interface regres-
sion. Moreover, we add random noise to the interface loca- IV. RESULTS
tion to simulate an experiment. This noise was uniformly dis-
tributed, with zero mean and 10µm standard deviation. The We begin the analysis by illustrating the evolution of the
parameter uncertainty was computed using the bootstrapping liquid height as a function of time on one side of the tube for
approach discussed in section III C. We compare this approach both liquids. Two representative examples are shown in Fig-
with the results obtained with standard tangent line method ure 9a for HFE7200 and Figure 9b for demineralized water.
(TLM). In this case, for each of the 30 generated profiles, we The height h̄ in these plots is the spatial average over the de-
perform a linear extrapolation of the interface using randomly tected menisci, shifted so that h = 0 corresponds to the equi-
sampled interface points among the last 100 µm to the wall. librium position. Each experiment starts with the same height
For each measurements we generate a Gaussian distribution of h̄ = 10 mm and the interface position oscillates within a
of contact angles by repeating the extrapolation ntlm = 500 range of ≈ 17.5 mm of axial distance within the tube. The
times using 70% of the available points. The contact angle is length of the fluid column controls the (natural) frequency of
obtained from the mean of the contact angle distribution, with the oscillation, which for the experiments presented here is
the uncertainty computed as 1.96 times the standard deviation. 2.5 Hz for HFE7200 (Figure 9a) and 2.7 Hz for demineral-
The results of this analysis are shown in Figure 8. This plots ized water (Figure 9b).
the contact angle derived via the MPM and TLM approach as a The gray area indicates the uncertainty on the average
function of the actual contact angle in the generated interfaces. height computed from the interface detection (see supple-
The results obtained using the MPM are plotted with circles, mentary material). These experiments were repeated three
with different markers filling distinguishing profiles with the times for repeatability analysis; since the standard deviation
three levels of interface resolution at the wall. The Figure also between various experiments is generally lower than the un-
 7




 

(a) Interface displacement for HFE7200 (b) Interface displacement for demineralized water

(c) Interface accelerations for HFE7200 (d) Interface accelerations for demineralized water

FIG. 9: Figures a-b show the evolution of the mean interface height h for a test case in HFE7200 and demineralized water. The
dashed line shows the response of a canonical second order system with ωn = 15.96 rad/s, ζ = 0.11 for HFE7200 and
ωn = 17.8 rad/s and ζ = 0.17 for water, where the zoom view highlights the stick-slip motion of the contact line. The
dashed-dot line shows the exponential envelope for the oscillation decay (e−λt ). Figures c-d show the acceleration of the spatial
averaged interface (solid line) and contact line (dashed). In the case of water (9d) the behaviour is chaotic because of the stick
slip of the contact line.

certainties in the interface detection within a single run, we The matching is overall satisfactory if one considers the sim-
hereby consider only one representative test case for each plification involved in modelling the interface dynamics using
fluid. For both fluids, the response of the liquid interface qual- an autonomous dynamical system.
itatively behaves as a second-order canonical linear dynamical
system of the form h̄¨ + 2ζ ωn h̄˙ + ωn2 h̄, which one could derive The higher viscosity and higher capillary forces in the case
from a simple macroscopic model for the liquid column dy- of water lead to 1.8 times higher decay rate. Moreover, in the
namics. case of water, the contact line systematically pins to the wall
a few instants before reaching the second peak. The snap-
The decay rate computed from the experimental data is also shots of Figure 10 show that this condition occurs when the
indicated in Figures 9a and 9b, and its exponential envelope rising interface meets the liquid film deposited at the previous
is shown with dash-dot lines. The dashed line illustrates the descend. Figures 10f-10h show that the film moves towards
prediction of the ‘best’ canonical second-order system for the the interface prior to the pinning, which is shown by Figure
data at hand, identified via regression of the natural pulsation 10i. After the pinning the velocity of the contact line is null,
ωn and damping factor ζ . The results of the regression are but the liquid column pushes upward the central part of the
indicated for each fluid in the captions of Figures 9a and 9b. interface as shown in Figure 10j. The associated ‘stick-slip’
 8

A. Experimental validation of Interface models

Here we analyze the experimental validation of the GB (3),

VG (5) and IC (7) models. The global performances of these
(a) 0.270s (b) 0.287s (c) 0.304s (d) 0.320s (e) 0.337s models are shown in Figures 11a and 11b for a tests case with
HFE7200 and demineralized water respectively. In these Fig-
ures, the p upper plots show the time evolution of the objective
function J(w∗ ) used to solve the regression problem, where
J(w∗ ) is evaluated at the optimal set of parameters w∗ , that is
w∗ = θ for p the GB and VG models and w∗ = ct , li , θ for the
(f) 0.270s (g) 0.287s (h) 0.304s (i) 0.320s (j) 0.337s IC model. J(w∗ ) is the root mean square of the (optimal)
Fréchèt distances41 between the experimental points and the
FIG. 10: Snapshots of the experiment with HFE7200 models’ predictions. The bottom plots show the time evolu-
(a-b-c-d-e) and demineralized water (f-g-h-i-j) in 8 mm tion of the Weber number We = ρv2i (2R)/σ , with vi = h̄˙ the
diameter tube. In 10i the rising interface encounters the velocity of the spatially averaged interface as described in the
descending film. The slope of the interface near the wall previous section.
suddenly increases and the interface stops its motion. Both figures should be analyzed together with figures 9a
and 9b. In the case of HFE7200 (Figure 11a), We ≈ 20 at the
first peak while it reaches at most We ≈ 4.5 for demineralized
water. During the first cycle (t < 0.2s), both the GB and the
motion of the contact line slows down the rising of the liquid VG models perform poorly –as expected– since the interface
interface (see annotation in Figure 9b and the zoomed view experiences large acceleration and oscillations in this phase.
around t ∈ [0.3 − 0.4]s). This dynamic is not observed in the For both models, the regression of the interface is harder in
case of HFE7200. the case of HFE7200 than for water (leading to larger errors);
The interface deformation and oscillations in these experi- we believe this is mostly due to the negligible contribution of
ments were observed to be fairly axial-symmetric. In the case surface tension in the first as compared to the second. Never-
of water, the film deposition on the wall alters the axisymme- theless, both GB (3) and VG (5) models produce satisfactory
try of the interface (cf. figures 10f-10g-10h) but this condition results at largest time, as velocity and acceleration vanish.
lasts only few milliseconds and does not appreciably impact The IC model (7) performs much better within the entire
the interface as shown by the figures 10. experiment. This is primarily due to the additional degrees of
freedom in the parameter space. While for both GB and VG
Two main mechanisms could be qualitatively identified.
models oscillations of the interface in the center of the channel
The first mechanism arguably originates at the wall: as the
must induce changes in the contact angle (and vice-versa), this
contact line velocity changes, the contact angle changes and
does not need to be the case for the IC. The additional term in
with it, at least up to some distance from the wall, the inter-
(6) has the main merit of decoupling the interface dynamics
face shape. The second mechanism originates at the channel’s
far from the wall from the dynamics of the interface near the
center: because the flow far from the wall moves faster, its
contact line.
acceleration involves larger inertial contributions. These trig-
gers interface oscillations, as shown in Figures 9c and 9d for We illustrate the importance of such a correction by
HFE7200 and water respectively. These Figures show the ac- analyzing the interface shape in three conditions of largely
celeration of the spatially averaged interface height h̄¨ (t) and different velocities and accelerations for the two fluids. These
the acceleration of the contact line (ḧ(R,t)). In both cases, are indicated as A, B, C in Figures 11a and Figure 11b. In
the time differentiation is carried out after smoothing the time the case of HFE7200, the spatial-averaged interface velocity
series using a Savitzky–Golay filter. (vi = h̄˙ ) and acceleration (ai = h̄¨ ) are (vi , ai ) =(-0.14 m/s,
-1.6 m/s2 ) in A, (vi , ai ) =(0.1 m/s, 0.08 m/s2 ) in B and
In the case of HFE7200 (Figure 9c), the interface oscil- (vi , ai )=(0.03 m/s, -0.01 m/s2 ) in C. In the case of deminer-
lates significantly at the center of the channel, at a frequency alized water, these are (vi , ai ) =(-0.17 m/s, -1.23 m/s2 ) in A,
(≈ 20Hz) much larger than the frequency of the liquid col- (vi , ai ) =(0.03 m/s, -0.01 m/s2 ) in B and (vi , ai ) =(-0.04 m/s,
umn’s oscillation. This oscillation begins right after the first -0.05 m/s2 ) in C.
minima and lasts until t = 0.5s. In the case of water (Figure
9d), the acceleration profiles are somewhat chaotic because of
Figure 12 shows the comparison of the interfaces models
the stick-slip motion of the contact line. Nevertheless, these
for the three conditions A-B-C and for each of the fluids. As
oscillations can be seen within the first 0.4s.
expected, the GB (Figures 12a-12b) and VG (Figure 12c-12d)
While in both fluids these oscillations are too small to be models are unable to describe the interface shape under strong
visible in the average interface height h̄(t) in Figure 9a and 9b, acceleration (case A). Moreover, even at low acceleration lev-
their impact on the contact angle is significant depending on els it appears that despite the low regression error the GB
which model is used for the interface regression, as discussed model overestimates the slope of the interface near the contact
in the following section. line and thus the contact angle. This is more evident for the
 9


JUDYLW\EDVHGPRGHO YLVFRXVJUDYLW\PRGHO LQHUWLDFRUUHFWHGPRGHO

J [ m]





We [ ]


$ % &

      
time [s]

(a) Regression error and Weber number history for the experiment with HFE7200

JUDYLW\EDVHGPRGHO YLVFRXVJUDYLW\PRGHO LQHUWLDFRUUHFWHGPRGHO
J [ m]






$ % &
We [ ]


      
time [s]

(b) Same as Figure 11a but considering a test case with demineralized water

FIG. 11: The Figure on the top shows the regression error as a function of time for the three models. The Figure on the bottom
shows the Weber number of the interface as a function of time and the three example conditions (A-B-C) at which the interface
is later analyzed in the what follows.

cases with HFE7200 (Figure 12a) than cases with water (Fig- ning at the wall while the rest of the interface is still advancing
ure 12b), which had a higher contact angle. The VG model (and subject to a large inertia).
performs satisfactorily for We < 5 (Figures 12c-12d), while Figures 13 and 14 compare the magnitude of each term in
for higher We (case B) this model also overestimates the slope the force balance of the IC model for both fluids. Each plot
of the interface near the contact line, especially in the case of shows the distribution of the absolute value for each term as
HFE7200. a function of the radial coordinate r. The title in each plot
Figures 12e-12f show the comparison of the same interfaces indicates the corresponding time (cf also Figures 9a-9b).
with the IC model. All the three conditions A-B-C are well These figures illustrate how different the force balance is in
described up to a close distance from the wall. In the case of these two fluids. Yet, the inertial contribution plays a leading
HFE7200 (Figure 12e), the experimental interfaces differ at role in all the illustrated snapshots, especially far from the
most 5% of the capillary length lc . wall and in the case of HFE7200. In the case of water, the
As pointed out in the previous section, the correction in the capillary force is the second most important contribution due
IC model is necessary to make the interface shape less depen- to the larger surface tension, while in the case of HFE7200 this
dant on the contact angle, and this is particularly important in is the viscous term. This reaches large values near the contact
receding conditions under large acceleration. The correction line because of the small contact angle, while it is negligible in
is effective in both fluids despite these feature largely different the case of water because of the much smaller interface slope
interface shapes. Particularly interesting is the test case B for at the wall.
water: this is taken at t ≈ 0.3s, when the contact line is pin- Interestingly, the shape of the inertial correction is similar
 10





 
   

(a) HFE7200 interface vs GB model (b) demineralized water interface vs GB model





 
   

(c) HFE7200 interface vs VG model (d) demineralized water interface vs VG model




 
   

(e) HFE7200 interface vs IC model (f) demineralized water interface vs IC model

FIG. 12: Comparison of the experimental interface data (circular markers) with the corresponding model regression (solid
lines) for three interface conditions (A-B-C). The plots on the left side correspond to an experiment with HFE7200; the plots on
the right corresponds to an experiment with demineralized water. From top to bottom, each row shows the regression using the
GB, the VG and the IC models respectively. The interface is shifted with respect to the mean value and normalized by the
capillary length lc . A zoom of the contact line region is shown on the top right each of each plot. At the largest acceleration and
receeding conditions, both the GB and VG models overestimate the contact angle especially in the case of HFE7200.
 11

V V V



0DJ>1@


 JUDYLW\
FDSLOODULW\
 YLVFRXVSUHVVXUH
LQHUWLD
V V V


0DJ>1@




              
U>PP@ U>PP@ U>PP@

FIG. 13: Profiles of the gravity (solid black line), capillary (dashed red line), viscous (dotted blue line), and inertia (dash-dot
green line) terms across the channel in the IC model for different time of an experiment, for the case of HFE7200. The inertial
term is much larger than the others during the first moments of the experiment.

V V V



0DJ>1@


 JUDYLW\
FDSLOODULW\
 YLVFRXVSUHVVXUH
LQHUWLD
V V V


0DJ>1@




              
U>PP@ U>PP@ U>PP@

FIG. 14: Same as Figure 13 but with demineralized water. In this case the capillary term is much larger and comparable with
the inertial term also in the first moments of the experiment.
 12

in the two cases, assuming a uniform value in the center of the Tanner law16 and the molecular-kinetic theory (MKT)9 . The
channel at the largest acceleration and reaching the saturation first express a linear trend in this plot with slope ≈ 12Ca−1
at the distance at which the capillary term reaches a maximum while the second simplifies to θ ≈ cos−1 (cos(θs ) − ζw/µCa)
(e.g. time 0.075 s and 0.16s in the case of water, 0.075 and with ζw = 12 mPa s. The MKT law has been written as a
0.45s in the case of HFE7200). function of the the capillary number and the contact line fric-
tion coefficient ζw 46 . Clearly, given that the capillary number
is of ∼ 10−3 the relationship between the capillary number
B. Transient contact angles and the dynamic contact angle is almost linear.
Although the relative uncertainties in this plot are consid-
Finally, we analyze the time evolution of the contact an- erable, due to the low contact angle, the trends suggest that
gle using the different interface models. Figure 15a and 15c such kind of correlations can be extended in the presence of
show the contact angle as a function of time for a case with significant acceleration and for the investigated system.
HFE7200 and demineralized water respectively. The data for The same is not true for the case of demineralized water,
each model contains only points for which the interface re- as shown in Figure 15d. No clear trend can be identified, and
√ the stick-slip motion might be only partially responsible for
gression gives J < 25µm. Hence, fewer points are available
for the GB and VG models in the first ≈ 0.5s. this since no relation can be found even at the largest Ca (i.e.
In the case of HFE7200 (Figure 15a), both GB and VG after the pinning has occurred and the contact line motion re-
models predict large oscillations of the contact angle in the established). The contact line pinning (leading to Ca = 0)
time interval t ∈ [0.2, 0.4]s. This is the time interval between is observed whenever the contact angle approaches θ = 60o .
the first minima and the first maxima (see also Figure 9a) and Whether this is the cause or the consequence of the pinning re-
is characterized by additional oscillations of the interface at mains unknown. Nevertheless, despite the excellent accuracy
the center of the channel (see also the oscillations in Figure of the IC model in the interface regression, no relation of the
9c). These models must rely on the contact angle (see section kind θ = f (Ca, A∗ ) can be identified since multiple contact
III) to accommodate the interface oscillation and thus both angles are observed for the same velocity-acceleration condi-
models tie the interface dynamic to the contact angle evolu- tions.
tion. But this link, in fact, is much weaker than what these
models predict.
The IC model, leveraging the correction term, makes the V. CONCLUSION
contact angle independent from these inertial additional oscil-
lations and predicts a smoother contact angle evolution. We This work analyzed the contact angle dynamics on low vis-
consider this more reliable as the corresponding interface re- cous liquids and in accelerating conditions, extending the lit-
gression has a much lower error (see section IV A). Similar re- erature on models and correlations for dynamic contact angles
sults are observed for the case of water (Figure 15c), although beyond the usual highly viscous liquids and steady contact
the larger surface tension helps the GB and VG models. Nev- line velocity. We consider two liquids: demineralized water
ertheless, as discussed in the previous section, these models (for its high surface tension) and HFE7200 (for its low con-
significantly over-predict the contact angle. tact angle).
We conclude the analysis by attempting to link the dy- Due to the high volatility of hydrofluoroethers and the low
namic contact angle to the velocity and acceleration of the accuracy of direct techniques (e.g. TLM), previous studies
contact line. The plots on the right of figure 15 map the dy- only focused on the static contact angle47,48 or assumed a con-
namic contact angles measured by means of the IC model to stant zero value in dynamic conditions49,50 . To the authors’
the velocity and acceleration of the contact line for an exper- knowledge, this study shows for the first time the wetting dy-
iment with HFE7200 (15b) and demineralized water (15d). namics of HFE7200.
The dimensionless contact line velocity is given by Ca(t) = We improve the dynamic contact angle characterization
ḣ(R,t)µ/σ while the dimensionless acceleration is scaled as by using an indirect Meniscus Profile Method (MPM) based
A∗ (t) = ḧ(R,t)/g. Both Ca and A∗ are given with their sign; on a model of the gas-liquid interface. We extend previ-
hence Ca < 0 corresponds to a receding interface while Ca > 0 ous approaches based on quasi-static (GB)25 and quasi-steady
corresponds to an advancing interface. (VG)26,27 interface models to the case of inertia-driven inter-
The marker are colored by the instantaneous dynamic con- face by proposing a heuristic ‘Inertia Corrected’ (IC) model.
tact angle θ (t) as predicted by the IC model. In the case of The results show that the IC model performed well for both in-
HFE7200, a clear trend is visible: the contact angle increases vestigated liquids over the entire range of conditions. In con-
towards large Ca regardless of A∗ . At Ca = 0 the contact an- trast, comparing the GB and VG models with the experimental
gle θ varies between 4.5◦ − 5.5◦ hence no appreciable wetting data showed that the coupling between the contact angle dy-
hysteresis is observed for HFE7200. This shows that classic namic and the interface dynamics at the center of the channel
contact angle correlations aiming at predicting θ as a function is much weaker than GB and VG models predict. The com-
of Ca can be identified. This is further illustrated in Figure 16, parison of the dynamic contact angle against standard models
which plots θ 3 (t) − θs3 as a function of the instantaneous cap- (i.e. not accounting for the acceleration) showed that in the
illary number Ca(t) and compares the experimental data with case of HFE7200 this could be linked to the instantaneous
the prediction of two well known correlations. These are the capillary number, as classic models would set, while for the
 13

[deg]




 

A* [ ]

 





      
Ca [ ] H 

(a) Contact angle evolution in HFE7200 (b) HFE7200 experiment

[deg]




 
A* [ ]

 






   
Ca [ ] H 

(c) Contact angle evolution in water (d) demineralized water experiment

FIG. 15: On the left: Evolution of the contact angle as a function of time according to the gravity based (GB),√viscous-gravity
(VG) and inertial (IC) models for an experiment with HFE7200 (15a) and water (15c). All regressions with J > 25µm are
excluded. On the right: scatter plot mapping all the snapshots in the interface evolution to the contact line velocity (in terms of
Ca) and acceleration (in terms of A∗ ). The markers are colored by the corresponding contact angle as predicted by the IC model
and the values are reported in the colorbar on the right of each plot.

case of demineralized water, this was impossible. An essential supported by the ESA Contract No. 4000129315/19/NL/MG
difference between the two configurations is that the contact and by ArcelorMittal’s ‘Minerva Contract’. Both partners are
line was found to pin multiple times in the case of water. gratefully acknowledged. Finally, the authors thanks Mathieu
Future work will focus on analyzing whether the results ob- Delsipee for its support and contribution in the preparation of
tained on HFE7200 hold for other highly wetting fluids and the experimental set up.
higher accelerations.

ACKNOWLEDGMENTS DATA AVAILABILITY

D. Fiorini is supported by Fonds Wetenschappelijk On- The data that support the findings of this study are available
derzoek (FWO), Project number 1S96120N. This work was on request from the corresponding author.
 14

17 T.-S. Jiang, O. Soo-Gun, and J. C. Slattery, Journal of Colloid and Interface

Science 69, 74 (1979).
18 M. Bracke, F. De Voeght, and P. Joos, in Trends in Colloid and Interface

Science III (Springer, 1989) pp. 142–149.

19 S. F. Kistler, Wettability 6, 311 (1993).
20 C.-L. Ting and M. Perlin, Journal of Fluid Mechanics 295, 263–300 (1995).
21 X. Bian, W. W. Schultz, and M. Perlin, Journal of Fluid
Dynamics (under revision), http://wwwpersonal. umich.
edu/schultz/Manuscripts/BianSchultzPerlin. pdf 146, 151 (2003).
22 D. Quéré, EPL (Europhysics Letters) 39, 533 (1997).
23 O. Shardt, P. R. Waghmare, J. Derksen, and S. K. Mitra, Rsc Advances 4,

14781 (2014).
24 G. R. Willmott, A. Briole, and F. Szczepaniak, Physical Review E 101,

043109 (2020).
25 J. G. Petrov and R. V. Sedev, Colloids and Surfaces A: Physicochemical

and Engineering Aspects 74, 233 (1993).

26 S. D. Iliev and N. C. Pesheva, Colloids and Surfaces A: Physicochemical

and Engineering Aspects 385, 144 (2011).

27 M. Maleki, E. Reyssat, D. Quéré, and R. Golestanian, Langmuir 23, 10116

(2007).
28 O. Voinov, Journal of colloid and interface science 226, 5 (2000).
FIG. 16: Comparison of dynamic contact angle data against 29 J. Eggers, Physics of Fluids 17, 082106 (2005).
30 P. Wu, A. D. Nikolov, and D. T. Wasan, Langmuir 33, 7862 (2017).
capillary number for an experiment with HFE7200. In the
31 M. Stange, M. E. Dreyer, and H. J. Rath, Physics of fluids 15, 2587 (2003).
case of receding contact line (negative Ca) the dynamic 32 M. H. Rausch, L. Kretschmer, S. Will, A. Leipertz, and A. P. Froba, Journal
contact angle seemingly saturate at its static value. In the of chemical & engineering data 60, 3759 (2015).
case of advancing contact line, the dynamic contact angle is 33 A. Buades, B. Coll, and J.-M. Morel, Image Processing On Line 1, 208
compared with Tanner law (solid red line), and (2011).
34 A. Savitzky and M. J. Golay, Analytical chemistry 36, 1627 (1964).
Molecular-Kinetic theory (solid blue line)
35 M. Mendez, L. Németh, and J.-M. Buchlin, in EPJ Web of Conferences,

Vol. 114 (EDP Sciences, 2016) p. 02072.

36 M. Darzi and C. Park, Physics of Fluids 29, 052004 (2017).
37 J. H. Snoeijer and B. Andreotti, Annual review of fluid mechanics 45, 269
REFERENCES
(2013).
38 L. Landau and E. Lifshitz, New York 61 (1959).
39 G. Delon, M. Fermigier, J. H. Snoeijer, and B. Andreotti, Journal of Fluid
1 T. Vestad, D. Marr, and J. Oakey, Journal of Micromechanics and Micro- Mechanics 604, 55 (2008).
40 J. Eggers, Physical review E 72, 061605 (2005).
engineering 14, 1503 (2004).
2 Y. Liu, A. Hansen, E. Block, N. R. Morrow, J. Squier, and J. Oakey, Journal 41 F. J. Mendez, A. Pasculli, M. A. Mendez, and N. Sciarra, Acta Geotechnica

of colloid and interface science 507, 234 (2017). 16, 2031 (2021).
3 K. Ma, R. Liontas, C. A. Conn, G. J. Hirasaki, and S. L. Biswal, Soft Matter 42 J. A. Nelder and R. Mead, The computer journal 7, 308 (1965).
43 P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cour-
8, 10669 (2012).
4 M. Kumar and A. Fogden, Langmuir 26, 4036 (2010). napeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der
5 G. R. Willmott, C. Neto, and S. C. Hendy, Soft Matter 7, 2357 (2011). Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson,
6 G. R. Willmott, C. Neto, and S. C. Hendy, Faraday discussions 146, 233 E. Jones, R. Kern, E. Larson, C. J. Carey, İ. Polat, Y. Feng, E. W. Moore,
(2010). J. VanderPlas, D. Laxalde, J. Perktold, R. Cimrman, I. Henriksen, E. A.
7 N. C. White and S. M. Troian, Physical Review Fluids 4, 054003 (2019). Quintero, C. R. Harris, A. M. Archibald, A. H. Ribeiro, F. Pedregosa, P. van
8 J.-C. Fernandez-Toledano, T. Blake, P. Lambert, and J. De Coninck, Ad- Mulbregt, and SciPy 1.0 Contributors, Nature Methods 17, 261 (2020).
44 U. M. Ascher, R. M. Mattheij, and R. D. Russell, Numerical solution of
vances in colloid and interface science 245, 102 (2017).
9 T. Blake and J. Haynes, Journal of colloid and interface science 30, 421 boundary value problems for ordinary differential equations (SIAM, 1995).
45 B. Efron and R. J. Tibshirani, An introduction to the bootstrap (CRC press,
(1969).
10 R. L. Hoffman, Journal of colloid and interface science 50, 228 (1975). 1994).
11 O. Voinov, Fluid dynamics 11, 714 (1976). 46 T. D. Blake, Journal of colloid and interface science 299, 1 (2006).
12 R. Cox, Journal of Fluid Mechanics 168, 169 (1986). 47 A. Sathyanarayana, P. Warrier, Y. Im, Y. Joshi, and A. S. Teja, Journal of
13 P. Petrov and I. Petrov, Langmuir 8, 1762 (1992). Nanotechnology in Engineering and Medicine 3 (2012).
14 Y. D. Shikhmurzaev, International Journal of Multiphase Flow 19, 589 48 Z. Cao, Z. Wu, A.-D. Pham, and B. Sundén, International Journal of Ther-

(1993). mal Sciences 146, 106107 (2019).

15 P. De Gennes, X. Hua, and P. Levinson, Journal of fluid mechanics 212, 55 49 N. Fries, K. Odic, M. Conrath, and M. Dreyer, Journal of colloid and inter-

(1990). face science 321, 118 (2008).

16 L. Tanner, Journal of Physics D: Applied Physics 12, 1473 (1979). 50 E. Fuhrmann and M. E. Dreyer, Experiments in fluids 55, 1 (2014).